FFFFF

ESRWF: Extreme State-Rank based Workload Factoring for Integrated Cloud Computing Model
Chapter-1
Introduction
There isnt person anywhere who capable of doing more than he thinks; he can. Yes. We
must become the change we want to see.
-Henry Ford, Founder of FORDS group of company, USA
1.1 Motivation
Figure 1.1: Cloud service & domain provider firms based on Cloud Computing
Cloud computing is a very current topic and the term has gained a lot of traction
being sported on advertisements all over the Internet from web space hosting providers,
through data centers to virtualization software providers. Cutting through the hype of
cloud computing is not an easy task as a simple web search suffices to convince that
there are nearly as many definitions on what constitutes 'cloud computing' as there are
Samrat Ashok Technological Institute, Vidisha
Page
players in the market seeking to gain new territory in that promising new business field.
IBM, Amazon, Microsoft, Google, Yahoo and Apple among others are very active in the
area of cloud computing. They either already provide cloud computing commercial
solutions in one form or another, or actively sponsor research centers, pursuing
development of marketable technology. Marketing rhetoric notwithstanding, the
academic world has also provided meaning and definitions on what does or should
cloud computing aim at and what are typical services that are expected to be
encompassed by the definition of cloud computing, as evidenced for instance by the
work of the RAD lab at the University of California at Berkeley or the GRIDS lab at the
University of Melbourne. Other renown companies such as Oracle are eager to provide
new offerings that allow enterprises to benefit from the developments taking place in
the area of Cloud Computing [29], yet they attempt to steer clear out of the hype and
highlight that they have redefined cloud computing to include everything that we
already do, as stated by Oracle's Larry Ellison. [30]
Furthermore, differently colored opinions on cloud computing technology from
industry speakers and experts ranging from praise and optimism to critique on the
viability and feasibility along with concerns on privacy, security and not least cost
efficiency of the currently offered cloud computing models are available as white papers
and seem to be broadly discussed within the IT community. Such complex technology
and business models setting entails an extensive research and provides the motivation
towards writing this bachelor thesis. The main goal is to clear the air on cloud
computing and provide an unbiased and independent, albeit critical outlook of the
technology. As the title of this thesis suggests its aim is to enable the reader to gain an
overview of the vital aspects of cloud computing in a three-fold way:
By providing common definitions of the important terms;

By setting apart the advantages of the technology and the disadvantages and
problems inherent to it; and

By ultimately delivering concrete technical and business model details on
popular cloud architectures, offered by the big players in the field.
1.2 Research Objective

Page
Special emphasis is put on the critical examination of each strategy as now more
than ever in the face of the global economic crisis, companies face higher refinancing
and investment costs and as any company thinking about adopting or moving to cloud
computing technology would do in practice, short-to-medium term disadvantages of the
technology have to be pragmatically and carefully weighted out against any hyped longterm potential efficiency achievements, be it strategic, technical or cost related.
Cloud Computing is an emerging trends of Distributed Computing and
generalization of Grid Computing known either as online services such as Amazon AWS
[12] & Google App Engine [13] for enabling convenient, on-demand services access to a
shared pool of configurable computing resources (e.g. Networks, Servers, Storage,
Applications, and Services) that can be rapidly provisioned and released with minimal
intelligent management efforts or service provider within interactions of regulations.
Where, virtualization is intelligent resource management facility of Cloud Computing as
service providence to Customer on support of Full-Virtualization as well as ParaVirtualization [21], [22]. Critics argue that Cloud Computing is not reliable enough due
to the tension created by incoming oppressive requests. It has features of shared
computing
infrastructure
for
hosting
multiple
applications
where
management
complexity is hidden and resource multiplexing leads to proficiency. More computing

resources are allocated on demand to application when its workload incurs more
resource demand than it is currently allocated. So, here we focus on Workload Factoring
[1] as solution in this paper as it is a unique functionality requirement by Cloud
Computing Architecture. Besides the presentation consolidation, we describe and
evaluate our technologies in context of video streaming (for experiments) and flow of
requests (for formulations) as applications throughout this white paper. We proposes an
Integrated Cloud Computing Model described in Figure 5.1 which enterprise IT
consultants can base to design and plan their computing platform for hosting Internet
based applications with highly dynamic workload. Our Integrated Cloud Computing
model features two analytical ideas as: ESRWF & ESRATE and also two states for
system which is naturally and randomly appeared different components in aggregate
workload of Internet applications.
Page
Figure 1.2 shows the dynamics of the hourly workload measured 1 during a 46days period on Yahoo! Video, the 2 nd largest U.S. online video sharing website. Applying
statistical analysis techniques including autoregressive integrated moving average
(ARIMA) and Fourier Transfer analysis, we observed that there were clear periodic
components (exampled by the workload between July 23 and July 27) in most of the
time; however, the big spikes shown in Figure 1.2 were not predictable. Therefore,
resource planning could be effective most of the time, but not always working We also
observed that the ratio of the maximum workload to the average load is as high as 5:3
(12:9 if workload was measured in half an hour interval), which makes over-provisioning
over peak load highly inefficient. Based on the above observations, we believe an
integration of proactive and reactive resource management schemes should be applied
on different components of the aggregated application workload along the time:
proactive management opportunistically achieves resource efficiency with reliable
prediction on the base workload seen most of the time, while reactive management
deterministically response to sudden workload surges in rare time with the requirement
of agile and responsive performance. [1]
Figure 1.2: Video stream workload evolution on Yahoo! Video Site (July-August, 2008)
Page
An efficient workload factoring philosophy is shaped for enabling technology

of the Integrated Cloud Computing Model. Its basic function is to split & control the
workload into two parts upon (unpredictable) load spikes, and assures that the regular
workload remains within planned in amount and the critical workload incurs minimal
cache/replication demand on the application data associated with it. This simplifies the
system architecture design for the Critical State and significantly increases the server
performance within it. As for the Regular State, workload dynamics are reduced
significantly; this makes possible resource capacity planning with low over-provisioning
factor and/or efficient dynamic provisioning with reliable workload prediction. We built
a video streaming service and flow of request for Formularize Description and Tri-Way
Trajectory Simulation as an experimental showcase of the Integrated Cloud Computing
Model. It has a local cluster as the regular workload region e.g. LDC [2] and the critical
workload region e.g. [12], [13]; the workload factoring scheme was implemented as a
load controller to arbitrate the stream load distribution between the two states. With
extensive
Analysis,
Observation
experiments, we showed
Tracery,
and
Tri-Way
Trajectory
Simulation
this workload factoring technology can enable reliable
workload prediction in the Regular State, effective ratio of the maximum load to the
average load, Memory-Prediction-Error () in Critical State, achieve resource proficiency
and reduce data cache/replication overhead in the Critical State, and react fast (with an
speedup factor) to the changing application data popularity upon the arrival of
load spikes.
1.3 Proposed Rationale

In the proposed Integrated Cloud Computing Model, which works for all cloud
deployment model with fair arranged integration of unlimited & limited both type
proposals
for
resolving
the
workload,
Model
defines two
states
for
resource
management:
Regular State which is original & has limited resources in dedicated application
platform in Local Internet Data Center (LDC) [2]. In this state all processes run all time
Page
and with balanced load of application & system. As processing the data volume doesnt
vary sudden and strike after removing at high workload. The LDC is expected to run in
relaxed mode and with high resource proficiency & managing facility for best utilizing
scenario even though resource provisioning for QoS guarantee.
Another is Critical State in which all resources in dedicated application platform
operated by cloud service provider & related infrastructure. This state is provisioned on
high demand and also optimizes to be on for a momentary variation of critical
oppressive runtime periods. Here, Resources are provided at large scale. Each state has
its local load balancing scheme managed according to workload present in runtime.
The origination of the Integrated Cloud Computing Model is based on dynamic
workload as Hub & Spoke Infrastructures [23]. It addresses many concepts where IT
consultants completely rely on extreme utilization of cloud services for application
hosting. At some cases, we cant predict some types of unexpected workload. It is
necessary to learn these natures of high spike & burst-time as well as find out an
efficient way to handle it once such events happen. Ongoing way of Customer Resource
Management (CRMs) [16] with Integrated Cloud Computing Model, we identified that the
resource management scheme applied through Integrated Cloud Computing Model has
efficient impact to resolve oppressive workload on different components of aggregated
applications during runtime as intelligent satisfaction for customers. This scheme
varies in our model as three ways
Opportunistically, achieves resource proficiency with reliable prediction on the
regular workload seen most of the time achieves through local data center.
Optimistically, response to high workload manages in rare time with the
requirement of agile and responsive performance through cloud services.

Deterministically, neglecting the response to sudden panic or oppressive
workload dispose in rare time when the system is not agile to give 100 percent
performance.
Page
Cloud User
IaaS
System Mentor
PaaS
SaaS
Cloud virtualization Platform
NaaS
Workload Factoring Scheme
Network Resource
Computing Resource
Storage Resource
Integrated Cloud Computing Model

Integrated Cloud Computing Model with Workload Factoring Scheme
Figure 1.3: Proposed Integrated Cloud Computing Model with Workload Factoring Scheme
In our Proposed Model Role of Information Life-Cycle Management (ILMs) [3] refers
to a large set of regulations, policies and strategies that handle Information-Regulation
& their storage system manually or automatically through computing resources
according to system state. ILMs underused through four strategies & their regulations
these are- Infrastructure, Policy, Operational Jobs and Definitions. Actually ILMs
comprises the policies, practices, regulations, SLAs associated applications and tools
used to approach the business value of information with cost consuming IT
infrastructure before its final disposition. It is count as part of EMC (Enterprise Content
Management) also. Through logical grouping, availability and dynamic nature of
information ILM proposed for Information Centric Approach. According to our
Rationale, ILM have important role to corroboration of reducing oppressive workload in
Critical State as following functionality-
Creation
&
Receipt
adopt
request
&
create
state
and
acknowledgement.
Distribution follows to categories the request according to states.
Page
process
with
Maintenance related to requests information privacy & security and their

regulations.
Uses refer to state handling resource management services.
Disposition offers non-accepted requests draining or their neglecting services.
Exception related to agreement discontinuity or non-recurring issues outside

normalcy.
Required Goodness for Request Estimation Scheme
It should extremely estimate request rate to specified accuracy as well as proved

their correctness above theoretical prediction. This implies that for a specified
accuracy. The shorter sampling time is better for every scheme. Extreme
estimation is critical for fast detection of anomalous events.

The scheme should be suitable for run time processing of traffic streams of
requested data items. This implies that the scheme performed should be simple
or preferably amenable to cover each step of implementations.

The scheme should be memory efficient practically. System may have millions of
data items and it should not be necessary to maintain system according to state
for large fraction of the data item for a few estimation results.
For schemes concreteness, we have to aimed design goals of workload factoring

solutions included four:
Reducing the higher workload and dynamic complexities in Regular State

oriented application platform and avoiding overloading scenarios through
Draining & Dispatching.

Making agile resource managed rationale for Integrated Cloud Computing Model.
Create Critical State oriented application platform enable through workload
decomposition based on preliminary features of empirical threshold of system.

Minimizing the data cache and replication overhead through Draining &
Dispatching of selected requests for similar data objects into Critical State.
Page
1.4 Dissertation Outline

The rest of dissertation is organized as follows: here Chapter-2 describes
Fundamentals & Literature Survey. Chapter-3 and Chapter-4 describe Problem
Statements and Proposed Works respectively. For evaluation, we have to Analysis and
Performance Criteria included in Chapter-5 Implementation. We conclude this paper
with future work in Chapter-6.
Page
Chapter-2
Fundamentals & Literature Survey
It is very easy to defeat someone. But it is very hard to win someone. So, always try to
win & never thinks about rest because your success is waiting for bright future specially.
-Dr. APJ Abdul Kalam, Ex-president of Republic India
2.1 Advent of Cloud Computing

The increased degree of connectivity and the increasing amount of data has led
many providers and in particular data centers to employ larger infrastructures with
dynamic load and access balancing. By distributing and replicating data across servers
on demand, resource utilization has been significantly improved. Similarly web server
hosts replicate images of relevant customers who requested a certain degree of
accessibility across multiple servers and route requests according to traffic load.
According to Wikipedia, the underlying concept of cloud computing can be dated
even further back to a public speech given by John McCarthy in 1961 where he
predicts that computer time-sharing may lead to the provisioning of computing
resources and applications as a utility. CONDOR Projects in 1988 gives Concept and
even technological approaches behind cloud computing can thus not be considered a
novelty as such and in particular data centers already employed methods to maintain
scalability and reliability to ensure availability of their hosted data. What is more, cloud
systems are, unlike e.g. grid computing, not driven by research first and then being
taken up by industry, but instead originates directly from commercial requirements and
solutions. It is rise to a small boom of cloud offerings which mostly consisted in a
rebranding of their existent in-house solutions and techniques, as well as a potential
Page
10
exposition of these capabilities to Consumers. Multiple new cloud domains and

providers have thus arisen and it is not surprising, that the term has found multiple
related, yet different meanings. In particular, the scope of areas and capabilities that
so-called clouds are applied for differs thereby strongly. The most typical representatives
for cloud related functionalities can currently be found in the following areas:
Data centers trying to maintain high scalability and increase availability;

Web server farms automating and stabilizing their servers, respectively the
users website;
In house attempts to balance resources over the business solutions;
External ASP-type offerings.
It must be made clear in this context that Clouds do generally not refer to a specific
technology or framework, but rather to a set of combined technologies, respectively a
paradigm / concept.
2.2 What is a Cloud?

Various definitions and interpretations of clouds and / or cloud computing
exist. With particular respect to the various usage scopes the term is employed to, we
will try to give a representative (as opposed to complete) set of definitions as
recommendation towards future usage in the cloud computing related research space.
In its broadest form, we can defineA 'cloud' is an elastic execution environment of resources involving multiple
stakeholders and providing a metered service at multiple granularities for a specified level
of quality (of service).
In other words, clouds as we understand them in the context of this document
are primarily platforms that allow execution in various forms across multiple resources
(and potentially across enterprise boundaries). All of which have in common that they
(directly or indirectly) enhance resources and services with additional capabilities
related to manageability, elasticity and system platform independency. To be more
specific, a cloud is a platform or infrastructure that enables execution of code (services,
Page
11
applications etc.), in a managed and elastic fashion, whereas managed means that
reliability according to pre-defined quality parameters is automatically ensured and
elastic implies that the resources are put to use according to actual current
requirements observing overarching requirement definitions implicitly, elasticity
includes both up- and downward scalability of resources and data, but also loadbalancing of data throughput. As shall be elaborated, future cloud systems should also
be able to maintain a pre-specified level of quality, respectively boundary conditions
(including performance, energy consumption, etc.) and should allow integration of
resources across organizational boundaries, integrating multiple stakeholders.
Figure 2.1: Cloud Computing Architecture based on SOMA (Service Oriented Modeling & Architecture)
Cloud computing is a emerging trends of Distributed Computing and

generalization of Grid Computing for enabling convenient, on-demand services access to
a shared pool of configurable computing resources (e.g. Networks, Servers, Storage,
Applications, and Services) that can be rapidly provisioned and released with minimal
intelligent management efforts or service provider within interaction regulations. This
cloud model promotes availability and is composed of three service models, and four
deployment models. However Cloud computing has no particular definition for
description even it is a difficult term to explain to most; even to technologists and IT
Page
12
professionals, the concept of computing in the cloud is a wide and generic in terms of
many specific areas within online and offline both types environment. The cloud is
defined as the Internet surrounding every part of our daily lives, similar to the clouds in
the sky. However many new enterprise related buzzwords have evolved from the original
computing in the cloud concept; While a common misconception for cloud computing
is merely workload on the Internet, the cloud offers many services, infrastructure
benefits and scalability which may not be possible within ordinary local data center.
When cloud storage is used as the primary location of files and documents, there is no
guarantee to certain trust is left in the hands of the storage provider to ensure certain
steps are taken to prevent data loss and maintain the integrity of the file system;
enabling maximum uptime, reducing downtime and sustain the highest levels of
physical protection and data security at oppressive workload.
Figure 2.2: Introduction to used Technology offered by Cloud Computing
2.3 Grid vs. Cloud Computing

In one of their earlier white papers on the topic IBM specifically highlight the
differences or rather the evolution of cloud computing over grid computing. [31] The
term grid computing denotes dividing a large task into many smaller ones that run on
Page
13
parallel servers. [31] Wikipedia provides a vague, citation-less definition broadly

describing grid computing as a form of distributed computing whereby a 'virtual super
computer' is composed of a cluster of networked, loosely coupled computers acting in
concert to carry out processor-intensive large tasks. RAD at Berkeley summarize
shortly, without elaborating further on the separation of the two concepts, that grid
computing suggests protocols that offer a form of shared computation over long
distances, but (contrary to cloud computing) those protocols and software solutions had
however not grown beyond their communities. [32] According to IBM, the key advantage
is that cloud computing not only is able to divide a large computational task into many
smaller tasks to run on parallel servers, but could also support non grid
environments, such as a three-tier web site architecture (i.e. separation of
presentation, application logic and database, e.g. the Model-View Controller (MVC)) that
runs standard or Web 2.0 applications. By that IBM probably stresses that large,
resource intensive community websites could be built and run efficiently upon cloud
architectures. Other researchers conclude that cloud computing is most likely to bring
about the advantages of grid computing as the single point of access for all the
computing needs of the customers.
Figure 2.3: Grid Computing Working Scenario
Page
14
Figure 2.4: Cloud Computing Working Scenario
2.4 Cloud Computing: Definition

In a survey conducted in July 2008, Cloud Computing Journal cites the
attempts to correctly define 'cloud computing' of 21 independent experts practitioners
and academics. [33] The opinions differ, but a pattern is found such that the wording in
almost all explanations hovers around the keywords scalability, on-demand, pay-as-yougo, self-configuration, self maintenance and Software as a Service. IBM takes a
technical stance and considers a 'cloud' to be a pool of virtualized resources that hosts
a variety of workloads, allows for a quick scale-out and deployment, provision of virtual
machines to physical machines, supports redundancy and self-recovery and could also
be monitored and rebalanced in real time. [31] A scientific definition is proposed by the
GRIDS Lab at the University of Melbourne:
"A Cloud is a type of parallel and distributed system consisting of a collection of
interconnected and virtualized computers that are dynamically provisioned and presented
as one or more unified computing resources based on service-level agreements
established through negotiation between the service provider and consumers. [34]
Page
15
The researchers thus emphasize that a 'cloud' is thereby not only a combination of
clusters and grids, but is also extended by the implied usage of virtualization
technologies such as Virtual Machines (VMs) to meet a specifically negotiated service
quality level. This definition implies and captures two potentially problematic issues of:
the business issue of negotiating the proper SLA from the customer's perspective
of having the technical capacity to correctly account for and
Guarantee the service outlined in that SLA at all (resource monitoring, failure
redundancy, rebalancing of workloads, etc. from the provider's perspective).
Hence, GRIDS' definition seems somewhat more neutral than Berkeley's one:
Cloud Computing refers to both the applications delivered as services over the
Internet and the hardware and systems software in the datacenters that provide those
services (Software as a Service - SaaS). The datacenter hardware and software is what
we will call a Cloud. When a Cloud is made available in a pay-as-you-go manner to the
public, we call it a Public Cloud; the service being sold is Utility Computing. [35]
Berkeley's
researchers
moreover
limit
their
definition
with
several
additional
assumptions
the ability to pay-as-you-go as the necessary billing model, implying Utility
Computing as inherent to Cloud Computing and

That users keep their data stored safely in the infrastructure whilst offloading
their problems to the SaaS Provider that offloads his problems to the Cloud
Computing Provider.
UC Berkeley's definition on cloud computing is firm with regard to zero capital

expenditure for taking advantage of cloud resources. This differentiation is important,
due to the apparent capital expenditures associated with making private cloud
environments publicly available. Berkeley's researchers propose a three way model for
provision and usage of cloud services, that could also be seen as recursive in case of
mash-up provider that is a cloud user of another platform at the same time:
Page
16
Cloud Provider SaaS Provider / Cloud User SaaS User

Berkeley
refrains
specifically
of
the
usage
of
the
terminology
Infrastructure/Hardware as a Service and Platform as a Service, which is commonly

found in cloud computing explanations by industry experts and academics rather
Utility Computing (used here again interchangeably with Cloud Computing) is classified
in three models Computation, Storage and Networking. In addition to IaaS and PaaS,
some experts [36] list the following building blocks of cloud computing:
Storage-as-a-Service
Database-as-a-Service
Information-as-a-Service
Process-as-a-Service
Application-as-a-Service
Integration-as-a-Service
Security-as-a-Service
Management/Governance-as-a-Service
Testing-as-a-Service
2.5 Essential Characteristics
Figure 2.5: Benefits through emerging trends of Cloud Computing
Page
17
On-demand self-service, a consumer can unilaterally provision computing

capabilities, such as server time and network storage, as needed automatically
without requiring human interaction with each services provider.

Broad network access, Capabilities are available over the network and accessed
through standard mechanisms that promote use by heterogeneous thin or thick
client platforms (e.g., mobile phones, laptops, and PDAs).

Resource pooling, The providers computing resources are pooled to serve
multiple consumers using a multi-tenant model, with different physical and
virtual resources dynamically assigned and reassigned according to consumer
demand. There is a sense of location independence in that the customer
generally has no control or knowledge over the exact location of the provided
resources but may be able to specify location at a higher level of abstraction
(e.g., country, state, or datacenter). Examples of resources include storage,
processing, memory, network bandwidth, and virtual machines.

Rapid elasticity, Capabilities can be rapidly and elastically provisioned, in some
cases automatically, to quickly scale out, and rapidly released to quickly scale
in. To the consumer, the capabilities available for provisioning often appear to be
unlimited and can be purchased in any quantity at any time.

Measured Service, Cloud systems automatically control and optimize resource
use by leveraging a metering capability1 at some level of abstraction appropriate
to the type of service (e.g., storage, processing, bandwidth, and active user
accounts). Resource usage can be monitored, controlled, and reported, providing
transparency for both the provider and consumer of the utilized service.
2.6 Service Models
Page
18
Figure 2.6: Cloud Service Models Domains & Interfaces
2.6.1 Cloud Infrastructure as a Service (IaaS)

The capability provided to the consumer is to provision processing, storage,
networks, and other fundamental computing resources where the consumer is able to
deploy and run arbitrary software, which can include operating systems and
applications. The consumer does not manage or control the underlying cloud
infrastructure but has control over operating systems; storage, deployed applications,
and possibly limited control of select networking components (e.g., host firewalls). It also
referred to as Resource Clouds, provide (managed and scalable) resources as services to
the user in other words, they basically provide enhanced virtualization capabilities.
Accordingly, different resources may be provided via a service interface: Data & Storage
Clouds deal with reliable access to data of potentially dynamic size, weighing resource
usage with access requirements and / or quality definition.
Examples: Amazon S3, SQL Azure.
Compute Clouds provide access to computational resources, i.e. CPUs. So far,
such low-level resources cannot really be exploited on their own, so that they are
typically exposed as part of a virtualized environment (not to be mixed with PaaS
Page
19
below), i.e. hypervisors. Compute Cloud Providers therefore typically offer the capability
to provide computing resources (i.e. raw access to resources unlike PaaS that offer full
software stacks to develop and build applications), typically virtualized, in which to
execute cloudified services and applications. IaaS (Infrastructure as a Service) offers
additional capabilities over a simple compute service.
Examples: Amazon EC2, Zimory, Elastichosts.
2.6.2 Cloud Platform as a Service (PaaS)

The capability provided to the consumer is to deploy onto the cloud
infrastructure consumer-created or acquired applications created using programming
languages and tools supported by the provider. The consumer does not manage or
control the underlying cloud infrastructure including network, servers, operating
systems, or storage, but has control over the deployed applications and possibly
application hosting environment configurations. It provides computational resources via
a platform upon which applications and services can be developed and hosted. PaaS
typically makes use of dedicated APIs to control the behavior of a server hosting engine
which executes and replicates the execution according to user requests (e.g. access
rate). As each provider exposes his / her own API according to the respective key
capabilities, applications developed for one specific cloud provider cannot be moved to
another cloud host there are however attempts to extend generic programming models
with cloud capabilities (such as MS Azure).
Examples: Force.com, Google App Engine, Windows Azure (Platform).
2.6.3 Cloud Software as a Service (SaaS)

The capability provided to the consumer is to use the providers applications
running on a cloud infrastructure. The applications are accessible from various client
devices through a thin client interface such as a web browser (e.g., web-based email).
The consumer does not manage or control the underlying cloud infrastructure including
network, servers, operating systems, storage, or even individual application capabilities,
Page
20
with the possible exception of limited user-specific application configuration settings.

also sometimes referred to as Service or Application Clouds are offering implementations
of specific business functions and business processes that are provided with specific
cloud capabilities, i.e. they provide applications / services using a cloud infrastructure
or platform, rather than providing cloud features themselves. Often, kind of standard
application software functionality is offered within a cloud.
Examples: Google Docs, Salesforce CRM, SAP Business by Design.
Overall, Cloud Computing is not restricted to Infrastructure / Platform /
Software as Service systems, even though it provides enhanced capabilities which act as
(vertical) enablers to these Systems. As such, I/P/SaaS can be considered specific
usage patterns for cloud systems which relate to models already approached by Grid,
Web Services etc. Cloud systems are a promising way to implement these models and
extend them further.
End users/ Subscribers/ Clients
Software as a Service (SaaS)
Platform as a Service (PaaS)
Infrastructure as a Service (IaaS)
Figure 2.7: Service Model Architecture of Cloud Computing
2.7 Deployment Models

2.7.1 Private cloud
Page
21
The cloud infrastructure is operated solely for an organization. It may be

managed by the organization or a third party and may exist on premise or off premise.
It is typically owned by the respective enterprise and / or leased. Functionalities are not
directly exposed to the customer, though in some cases services with cloud enhanced
features may be offered this is similar to (Cloud) Software as a Service from the
customer point of view.
Example: eBay.
2.7.2 Community cloud

The cloud infrastructure is shared by several organizations and supports a
specific community that has shared concerns (e.g., mission, security requirements,
policy, and compliance considerations). It may be managed by the organizations or a
third party and may exist on premise or off premise. Typically cloud systems are
restricted to the local infrastructure, i.e. providers of public clouds offer their own
infrastructure
to
customers.
Though
the
provider
could
actually
resell
the
infrastructure of another provider, clouds do not aggregate infrastructures to build up

larger, cross-boundary structures. In particular smaller SMEs could profit from
community clouds to which different entities contribute with their respective (smaller)
infrastructure. Community clouds can either aggregate public clouds or dedicated
resource infrastructures. We may thereby distinguish between private and public
community clouds. For example smaller organizations may come together only to pool
their resources for building a private community cloud. As opposed to this, resellers
such as Zimory may pool cloud resources from different providers and resell them.
2.7.3 Public cloud

The cloud infrastructure is made available to the general public or a large
industry group and is owned by an organization selling cloud services. Enterprises may
use cloud functionality from others, respectively offer their own Services to users
outside of the company. Providing the user with the actual capability to exploit the
Page
22
cloud features for his / her own purposes also allows other enterprises to outsource
their services to such cloud providers, thus reducing costs and effort to build up their
own infrastructure. As noted in the context of cloud types, the scope of functionalities
thereby may different.
Example: Amazon, Google Apps, Windows Azure.
2.7.4 Hybrid cloud

The cloud infrastructure is a composition of two or more clouds (private,
community, or public) that remain unique entities but are bound together by
standardized or proprietary technology that enables data and application portability
(e.g., cloud bursting for load balancing between clouds).Though public clouds allow
enterprises to outsource parts of their infrastructure to cloud providers, they at the
same time would lose control over the resources and the distribution / management of
code and data. In some cases, this is not desired by the respective enterprise. Hybrid
clouds consist of a mixed employment of private and public cloud infrastructures so as
to achieve a maximum of cost reduction through outsourcing whilst maintaining the
desired degree of control over e.g. sensitive data by employing local private clouds.
There are not many hybrid clouds actually in use today, though initial initiatives such
as the one by IBM and Juniper already introduce base technologies for their realization.
Page
23
Figure 2.8: Integrated View of Cloud Deployment Models (Private, Public, Hybrid)
2.7.5 Special Purpose Clouds

In particular IaaS clouds originating from data centre have a general Purpose
appeal to them, as their according capabilities can be equally used for a wide scope of
use cases and customer types. As opposed to this, PaaS clouds tend to provide
functionalities more specialized to specific use cases, which should not be confused
with proprietariness of the platform: specialization implies providing additional, use
case specific methods, whilst proprietary data implies that structure of data and
interface are specific to the provider. Specialized functionalities are provided e.g. by the
Google App Engine which provides specific capabilities dedicated to distributed
document management. Similar to general service provisioning (web based or not), it
can be expected that future systems will provide even more specialized capabilities to
attract individual user areas, due to competition, customer demand and available
expertise. Special Purpose Clouds are just extensions of normal cloud systems to
provide additional, dedicated capabilities. The basis of such development is already
visible.
2.8 Current View

Page
24
Figure 2.9: Working Profiles of Workload Factoring at Hybrid Cloud Deployment Model
Critics argue that Cloud Computing is not secure enough due to the tension created
by incoming of oppressive requests. Cloud Computing has features of shared computing
infrastructure hosting multiple applications where management complexity is hidden
and resource multiplexing leads to proficiency. More computing resources are allocated
on demand to application when its workload incurs more resource demand than it is
currently allocated. We focus on Workload Factoring [1] as solution in this paper as it is
a unique functionality requirement by Integrated Cloud Computing architecture.
Besides the presentation consolidation, we describe and evaluate our technologies in
context of video streaming and flow of information as applications throughout this white
paper. This paper proposes an Integrated Cloud Computing model which enterprise IT
consultants can base to design and plan their computing platform for hosting Internet
based applications with highly dynamic workload. The cloud computing model features
two analytical ideas as: ESRWF & ESRATE and also two states for system which is
naturally and randomly appeared different components in aggregate workload of
Page
25
Internet applications: Regular State and Critical State are explicitly arranged in separate
CRMs, SLAs & ILMs with consistory of management.
Figure 2.10: Working Profiles of Load balancing at Hybrid Cloud Deployment Model
An excellent workload factoring mechanism is designed as an enabling

technology of the deployment cloud computing model. Its basic function is to split the
workload into efficient parts upon (unpredictable) load spikes, and assures that the
regular load part remains within planned in amount, and the critical load part incurs
minimal cache/replication demand on the application data associated with it. This
simplifies the system architecture design for the critical load zone and significantly
increases the server performance within it. As for the regular load zone, workload
dynamics are reduced significantly; this makes possible capacity planning with low
over-provisioning factor and/or efficient dynamic provisioning with reliable workload
prediction. Two years ago an Intelligent Workload Factoring (IWF) algorithm [1] built a
Page
26
video streaming service testbed as a showcase of the hybrid cloud computing model. It
has a local cluster as the regular load resource zone and the Amazon EC2
infrastructure [1] as the critical zone; the workload factoring scheme was implemented
as a load controller to arbitrate the stream load distribution between the two zones.
With extensive analysis, trace-driven simulations, and testbed experiments, IWF showed
the workload factoring technology can enable reliable workload prediction in the base
load zone (with simple statistical technique), achieve resource efficiency (e.g., 78%
higher server capacity than that in base load zone) and reduce data cache/replication
overhead (up to two orders of magnitude) in the trespassing load zone, and react fast
(with an
speedup factor) to the changing application data popularity upon the
arrival of load spikes.
Page
27
Chapter-3
Problem Statement
If friendship is your weakest point, then you are the strongest person in the world. Its one
of the life-long achievement & inspirable moment.
Abraham Lincoln, Ex-President, USA
3.1 Problems Specification Premises
Abuse and Nefarious use of Cloud Computing.

Insecure Interfaces and APIs.
Malicious Insiders.
Shared technology Issues.
Data Loss, Leakage and De-duplication.
Account or Service Hijacking.
Unknown Risk Profile
Efficient Cryptography approach against ordinary mechanisms with best
network utilization.
Authentication and Protocol for Services for secure trusting.
Cheap data and data analysis.
Cost-effective defense of availability.
Increased authentication demands.
Mash-up authorization.
Green Cloud Computing Mechanisms.
Work Overloading
3.2 Problem Identifications
Data De-Duplication & Trespassing

Efficient Authentication Management & SLAs
Large Cloud Storage Space
Data Hashing & Storing as making cluster
Data Partitioning & Workload factoring
Page
28
Data intelligence (Information-Centric)

Estimation Cost of data transmission
Workload Factoring
3.3 Statements about Identified Problems

Here, we instruct about above mentioned problems at identification points. In
this we say that there are few problems that abbreviate cloud storage facilities to make
most awaited techniques of green cloud computing. As example in India, at time of ICC
Cricket World Cup (ODI & T-20 both type format) organizes in every two-four years of
interval then workload on cricket service provider websites (such as espncricinfo.com)
have high spikes due to high workload. During this time these sites are not reliable give
their average performance to user. For increase such reliability of our logical
inspirations, in mentioned problems we describe data de-duplication that is about of
storage of duplicate data as non-scalable criteria. Which is one of the causes of large
wastage data storage; due to this fault we introduce ineffective data partitioning and
their operated workload factoring.
Data intelligence refers to automation about given instruction according to SLAs,
Authentication
Services,
Ruled Databases
and
their
excellent
performance
at
utilization. Now a time we miss such type of any approach as well as their maintenance
& storage. These are general description of problem but at networking level processing,
importance of such problem is increase in absence of any recommended technology.
3.4 Intelligent Workload Factoring (IWF)

An intelligent workload factoring service is designed for Hybrid Computing
Model. It enables federation between on- and off- premises infrastructure for hosting an
Internet based application and intelligence lies in the explicit segregation of base
workload and trespassing workload, the two naturally different components in
application workload. The core technology of intelligent workload factoring service is fast
Page
29
frequent data item detection algorithm, which enables factoring incoming request not
only on volume but also on data content, upon changing application data popularity.
Application Request
Base Load Thresho
Base Zone
Trespassi
ng Zone
Workload Profiling
Fast factoring
Figure 3.1: Inspirational Logic View of workload factoring for IWF
There are fast partition solutions proposed like the bi-section partition scheme
[11]. For video streaming services where request-data relationship is simple and there is
no net cut as one request accesses only one data item, the partition problem
degenerates to the knapsack problem where our greedy scheme is moving vertices from
the base zone one by one ranked by their popularity until reaching the trespassing
zone's capacity. This is equal to redirect the requests for the most popular data items in
a top-k list into the trespassing zone, and the remaining question is on how to quickly
generate the correct top-k list during a popularity transition time disturbed by the
workload burst.
Page
30
Figure 3.2: Implemented View of FastTop-K algorithm for IWF
IWF call the fast frequent data item detection algorithm FastTop-K. As shown in
Figure 3.2, it has the following data structures: a FIFO queue to record the last c
requests, a list to record the current top-k popular data items, a list to record the
historical top-k popular data items, and a list of counters to record the data items
access frequency. Given a request r, the algorithm outputs base if r will go to the base
zone otherwise trespassing. The key ideas in the fastTop-K algorithm for speeding up
frequent data item detection have two: speeding up the top-k detection at changing data
popularity distributions by pre-filtering old popular data items in a new distribution,
and speeding up the top-k detection at a data popularity distribution by pre-filtering
unpopular data items in this distribution.
3.5 Proposed Workload Factorings Problem

We model the Workload Factoring through the fast partitioning schemes like the
bi-section partition scheme [4]. Assume, each object such as images, video clips, any
database table or URL links also in the application data is modeled as Vertex, each
request type of this application data items named as Net and a connection of access
the relationship between a Net and Vertex is called its Pins. Now follow the guidelines
Page
31
of Multilevel K-Way Hypergraph Partitioning an NP-Hard problem i.e.[4] is to assign all

vertices data objects to K (K=3 in our case) disjoint non empty locations without the
expected workload beyond their capacities. In general the partitioning objective is to
minimize Net-Partition-Cost.
So, by definition of a Hypergraph

and a set of nets
size of a net
C( n j )
nj
N . Each net
is associated with each net
its nets i.e.
consist of a set of vertices
connects a subset of vertices in
is equal to the number of its pins i.e.
its nets and denoted as Nets (
v i V
n j N
H=(V , N )
D ( v i )=|Nets(v i)| . The weight
W ( v i)
vi
V . The
S ( n j )=|Pins (n j)| . A cost
n j . The nets connecting vertex
v i ). The degree of vertex
vi
are called
is equal to the number of
is associated with each vertex
is the expected workload caused by this data item, which is calculated as its
popularity multiplied by average workload per request to access or process data. The
cost
C( n j )
of a net
n j N
is expected data access overhead caused by this
request type, which is calculated as the expected number of request of this type
multiplied by the sum of its neighboring data objects size.
{v 1 , v 2 , . , v k }
vk
is a K-way vertex partitioning if each part
empty. Parts are pair wise disjoint and the union of parts gives
V . In
is non-
, a net is
said to connect a part if it has at least one pin in that part. The Connectivity Set
of a net
nj
is the set of parts connected by
Page
32
nj
. The connectivity
j=| j|
nj
of a net
is equal to the number of parts connected by
n j is an internal net. If
nj
=1 , then
. If
j> 1 then n j is an external net and is said to be Cut.
The K-way Hypergraph partitioning problem is defined as finding a vertex
{v 1 , v 2 , , v k }
partition
for a given Hypergraph
H=(V , N )
such that a
partitioning objective defined over the nets is optimized while a partitioning constrained
is maintained. The K-way Hypergraph partition, an NP-Hard problem [4] is to assign all
vertices data objects to K (K=3 in our case) disjoint non empty locations without the
expected workload beyond their capacities. In general the partitioning objective is to
minimize
cost
function
defined
CutSize ( )= C (n j) , let another weights
over
S i1
the
and
Net-Cut-Cost.
S i2
of a vertex
Then
v i V
is
data size of this objects then achieve minimal partition cost or Net-Partition-Cost-
Min
n j N cut
C j+ j
vi V critic al
S1
v i V criticaldrain
S2
}]
Where,
n j N cut
Cj
, Net-Cut-Cost defines total weights of the nets that span more than one
location, therefore bringing remote data access/ consistency overhead.
v i V critical
S i1
v i V critical drain
Si2
, is the total size of data objects in critical state and
practically represents the data transfer/ replication overhead.
j , is a factor to assign different weights on two overhead components.
Page
33
i
i
S 1 and S 2 another weights of a vertex v i V
is data size of this objects
We identified through our initial simulation process, for video streaming service
or flow of requests (for such as image, audio, PDFs) where request-data relationship has
no Net-Partition-Cost. Yes, to make effective partitioning technique we move on
System-State Partitioning approach from [4] where requests in Regular State one by one
ranked by their popularity until reaching the Empirical Threshold & CN-46 follow-ship
of the system to dispatch in the Critical State processing. This is equivalent to redirect
the requests of higher popularity revenuer data item have State-Rank 1 or in
ESRATE[5] into the Critical State. Then question arises that how can we generate
efficient State-Rank at continue runtime with high workload during popularity revenue
measurement.
3.6 Related Works

3.6.1 For Workload Factoring Algorithm
Amazon launches CloudFront [24] for its AWS customers (who can now deliver
application load through global network) in November 2008 and after two months, in
February 2009, H. Zhang, G. Jiang, K. Yoshihira, H. Chen, and A. Saxena proposed
Intelligent workload factoring for a hybrid Cloud Computing model [1] with
introduction to base workload zone and trespassing workload zone with formularized
descriptions of Traffic-Estimation of [6], [7], [18]. Around the same time period VMWare
also release the vCloud service concept [25] for Virtual Data Center Operation System
as experimental view of Para-Virtualization, which helps cloud customers to leverage
off-premise computing capacity. On follow this way, Berkeley researchers [19] offer a
ranked list of obstacles to the growth of Cloud Computing. Similar to the points, we
made in the introduction, the concerns on public Cloud Computing services include
service availability, data confidentiality and auditability, performance unpredictability,
and so on. On discuss SOMA (Service Oriented Model & Architecture), we believe an
Integrated Cloud Computing Model makes sense to enterprise IT and can resolve many
Page
34
issues raised from Integrated Cloud Computing model. Through Content Distributed
Network (CDN) and web caching workload factoring happen between a local web server
and proxy servers. The typical method is DNS redirecting and the workload factoring
decision is predefined manually over a set of most popular web objects. When current IT
systems evolve from the dedicated platform model to the shared platform model along
the cloud computing trend, we believe a core technology component is required on
flexible workload management working for such models mentioned above, and our
workload factoring technology (ESRWF) is proposed as one answer for it.
3.6.2 For Workload Estimation Algorithm

The work of [43] presented a sampling method that first selects the oppressive
objects, i.e., requests with rate above a certain threshold (say, 1% of the entire traffic),
and then counts all data items belonging to these oppressive objects. For this scheme,
deriving the number of samples needed to achieve specified estimation accuracy does
not appear to be easy.
A white paper [40] proposed RATE, a flow estimation mechanism based on
counting two-runs (flow id matches for two consecutive samples from the traffic
stream). For a given accuracy level, RATE requires worst-case sampling time slightly
longer than (1.38 times) the naive counting scheme. However, it uses significantly less
memory (square root of number of samples). A drawback is that for the same sampling
time, RATE has much worse accuracy than naive counting for flows with low rates.
ACCEL-RATE [38] was proposed as an enhancement to the original RATE scheme. In
ACCEL-RATE, each arrival packet is hashed into multiple buckets based on its flow id,
so that packets of the same flow are more concentrated" within each bucket. As a
result, a larger number of two-run samples can be generated for each flow. It is shown
that under uniform hashing the sampling time, in comparison to RATE, can be reduced
to 7:3k where k is the number of buckets. This is achieved with about 2/7 times more
memory needs than RATE. Note that, however, such sampling time reduction only holds
when hashing is uniform, which implies the rate of the largest request, should be
significantly less than 1k. Hence ACCEL-RATE is best suited for cases where the
Page
35
maximum flow rate is small and a loose upper bound is known a priori. In general, all
the proposed mechanisms tend to trade reduced estimation accuracy and/or increased
sampling time for lower memory cost. For a given sampling time, nave counting scheme
can still produce the most accurate results regardless of flow rate but at the expense of
much higher memory requirements. Other side, CATE [7] extends RATE with using the
width of the coincidence interval of the naive counting scheme as the benchmark to
determine the effectiveness of the sampling schemes developed. Fast Top-K [1] is the
first paper which contributes request patterns instead of flow mechanism with fully
inspiration of CATE. Follow same ideals of Fast Top-K, here we use State-Rank
terminology to access rate-estimation as well as efficient rate-concentration with
memory usage also.
Page
36
Chapter-4
Proposed Works
Gravitation is not being held responsible for people falling in love. It is only a matter to
which science or any technology cant reach.
-Albert Einstein. Two times NOBEL Prizes honored Great Scientist
4.1 Inspirational Logic View

We give logical inspiration through fig.2 for ESRWF (Extreme State-Rank based
Workload Factoring) scheme which has four basic components: System Mentor, StateResource & Workload Manager
Empirical Resource Estimator
Threshold Indicator
, Drain
State-Rank Distributor
RSystem Mentor
1 Migration 1, Cloud
0 Controller
0, LDC
Rank Distributor, Migration Controller and Resource & Workload Manager. The System
Mentor updates the information of currently running system in case of workloads upon
incoming requests and also keeps track of system performance and response to request
of each data item. State-Rank Distributor distributes state-rank to each request as 0 (for
Regular State) & 1 (for Critical State) to follow empirical threshold of empirical
threshold indicator.
Figure 4.1: Inspirational Logic View of ESRWF with their elementary parts
Resource & Workload Manager has two components these are Empirical Resource
Estimator & Threshold Indicator may be say as primary & secondary component also
Page
37
respectively. The primary component checks the system state and estimates the request
rate and request weight coefficient. This specifies the system performance & capacity,
availability of resources and maximum loads that can be handled by system in Regular
State. It may be set through previous information of system history that may be
managed through predefined instructions with automatically or may be manually
operated based on average records of previous Regular State which are used as resource
provision decision. And secondary component of Resource & Workload Manager is
based on empirical threshold of system is only and only applied to request has staterank 1 in Critical State, it may be considered as anomalous entity so it redirect to
cloud services otherwise drained for reliable & secure system from high unacceptable
workload.
After tagging state-rank and designation of threshold as 0, 1 in which 1 stateranked request again ranked by Migration Controller for redirecting or draining. If it
overhead the empirical threshold and MWL than it must be processed to drain with
state-rank updated as . So, Migration Controller takes decision according to
prerequisite conditions that which request type or rank type be redirecting to LDC or
Cloud Services or Drained.
4.2 ESRATE Algorithm

In this dissertation, we address the problem of accurate measurement of traffic
in a Cloud Computing Cloud. Measurement of traffic is an important component for
resource planning & management, workload factoring, detecting DoS attacks, and in
traffic engineering [36], [37]. The traffic in the cloud can typically be classified into
request and measurements are required on a per-request basis for particular service
requirement. The definition of request can be very flexible. Examples are specific
application to application traffic characterized by State-Rank, all traffic destined
toward destination cloud. The standard approach used for measuring traffic is to
sample the traffic arriving at the server, keep counts of the traffic arrivals on a perrequest basis and then use this Counter & State-Rank to estimate the traffic. The main
problem with this approach is scalability.
Page
38
Now, we process algorithm ESRATE: Extreme State-Rank bAsed Traffic

Estimation, The new scheme works by keeping registers for k previous arrivals and
comparing the new arrival with each of them with attesting State-Rank according to
system state. Matching between the new arrived request and one of the previous arrived
requests in registers are updated weight and State-Rank with follow system state. Now,
we process algorithm, likely other estimation mechanism, which maintains two tables:
Stateless Predecessor Table (SPT) maintains the first request which have no
identical states since pick in CSRT to the system. It is FIFO based primary table
& filled directly or by filtering as form of system buffer.

Count & State-Rank Table (CSRT) where each arrived request into the system,
through matching if the incoming request
of data item
F , which already
in CSRT then increase their count by 1 and update state-rank according to

system state otherwise apply filtering approach. If match is done in FIFO then
Nidus counts the request in CSRT according to algorithm.
Given a request R, the algorithm outputs are Regular State processing if r will
Comparison
Counter
Counting
Stateless
Predecessor
Table
Data-ID Weight S-R

Request
Filtering
R R1 R2
+
Matching
R
Count and State-Rank Table
the Regular State with state-rank 0 otherwise Critical State Processing

go to
state-rank
1 otherwise drained from this approach. It works as followed [5]:
if it has
Figure 4.2: ESRATE Nidus View and Implementation of algorithm
If the system is in Regular State, the CSRT (Count & State-Rank Table) is always
set as Empty when first request R is arrived.
Page
39
If R matches any data item of CSRT (for asking same data) increase the counter
of data item by 1 in counter column and update state-rank 0.

If system is setting as Regular State after going Critical State then on basis of
MWL (Maximal Workload Left) [19], request is processed by LDC.

Otherwise, randomly draw M requests from FIFO queue named as SPT (Stateless
Predecessor Table) & compare them with R, if R matches any of the M request
(for asking same data), pick that data item & put in CSRT with initialization of
counter as 1 and state-rank 0 & update CSRT.

If any request doesnt arrived again then system automatically dispose request
from SPT, because there is no mean of that request to keep continue in SPT.
At starting, if any data item requested at Critical State then it follow same
procedure as when it is entered in Regular State. If data item of request R
doesnt belong to CSRT then add R into SPT FIFO queue for request logs &
returns.
In Critical State, reset all counters to 0 and set state-rank 1 and calculating
the request rate of each data item participated in CSRT. For each data item in
CSRT, the request rate is its counter value divided by total requests arrives since
entering the system in Critical State. Also, calculate estimated request rate
correspond of each data item according to Section-V.

If requests of data item cross or overhead the Empirical Threshold Indicator
means workload so high not managed by system then request will be signified
with state rank since draining it automatically. Otherwise redirect to cloud
services on basis of LWL (Least Workload Left) [19]as prerequisite Workload
Factoring scheme with increment count by 1 and state-rank 1.
The reason for the calculation of the request weight coefficient with
request rate in cloud is that, the user feels request behavior inflicts two types of request
rates: Peak Request Rate (20 videos per 6 seconds) which discriminates with less
request weight coefficient & LWL basis to cloud services, other is Average Request Rate
(20 videos per 60 seconds), the ARR is a long term rate while the PRR is a temporal
rate have MWL be undertaken by LDC.
Page
40
Figure 4.3: Implemented View of ESRWF with their elementary parts
4.3 Controversial NonEntityship-46

CN-46 is our new proposal to approach draining of noxious requests of high
spikes from running system. When system is in Critical State and these requests may
become noxious to bring high workload for the system if it will be continue to process at
LDC or any cloud services. NonEntityship-46 defines that a request through which
services are provided to Customer, is adopted as entity for system and give performance
under-rule of or may say as it follows systems empirical threshold limitation means
Non-Entity, through average performance of requests, Resource & Workload manager
designs unique structures of 6 metrics for Entity-Ship, if any request overhead the basic
threshold value for 6 metrics that will be considered as controversial entity for system.
Than that System mentor alerted to system and after it, if this request again served
than that request with previous same condition will be drained. These requests are
handled by reserved pool of servers named as Drain Resolver works for controlling &
managing workload at Critical State. These servers will be free from continuous
internetworking process, activated at time of high workload only.
Page
41
This 6 metrics followed according to systems basics configuration & their

performance in critical issues of workload used in prime time services of internet but
required that the requests have follow any 4 metrics. If any request follows CN-46
means it crosses all empirical threshold limitations of states must be drained. If any
request doesnt follow any 4 metrics under rule of CN-46 than system mentor alerted
the system for that request with assign state-rank . If this request again requested
then without concern on 46 rules, it directly drained from system without checking of
NonEntityship-46 & dispatch to Drain Resolver. Otherwise redirect to cloud services
with State-Rank 1. A few Metrics for CN-46:
Comparison of actual system workload and predicted system load through
upcoming requests.
Unwanted Time-out conditions.
Resource Availability, Speedup and their proficiency.
Data cache and Replication overhead/ Consistency in Critical State.
Estimation Time, Memory Size, Bandwidth Allocation.
Security and Limitation of authorization levels.
Prediction-Errors and Strong SLAs & ILMs.
Data item Popularity and arrivals of load spikes.
4.4 Notations
We assume that each request belongs one of
denoted by
Pf =
rf
rf
= r f
fF
data item. The rate of request
denote total request rate to system. Let
denote the proportion of request rate or actual request rate to the system
that belongs to request
rf
and let
for each
f F . We have to design an efficient scheme to estimate
f F . Since it is easy to measure , instead of directly estimating
r f , we solve the equivalent problem of getting an estimate
Page
42
^
Pf
of
Pf
for each
^
Pf
f F . Then we use
r f . We can view
to estimate
that an arriving requests belong to data item
Pf
F . We assume that
as the probability
Pf
is static or
stationary over the time in which the estimation is done. We also assume that the
probability that an arriving request belongs to a given data item is independent of all
other requests. We can sample randomly in order to reduce this dependence. We now
give the accuracy requirement for ESRATE-
^
Pf
ESRATE will determine an estimated
such that-
^
Pf Pf , P f + if P f
2
2
^
P f Pf , P f +
if Pf >
2
2
With probability greater than
. In other words, we are willing to tolerate an error of
with probability less than
probability less than
for all
for all
and parameter
and an error of
with
Pf > . We consider that the proportion for most
data item lies below some threshold proportion

accurate in the range [0,
Pf
and we want the estimation to be
]. Formally, we are given threshold proportion
0 1
1 . A case, if there are data items with proportion greater than
, we still want the estimation to have a guaranteed performance but we are willing
to sacrifice the quality of guarantee somewhat. The ESRATE will estimate
Page
43
Pf
with
relative estimation error range of
[ ]
1
,1
2
for unit normal distribution, such as if
. We use
=99.9
to denote the
then
Z =4.00 .
percentile
4.5 Main Results

As per follow complete process of [7].
Lemma 1: Given the modified accuracy requirement, let N be the number of request
samples required by ESRATE. If the accuracy requirement for large data item can be
relaxed to-
2
1
+1
2
3 +1
( )
( )
Then setting,
Proof: Let
and
1 1
k = ( +1)
2
be the desired estimation accuracy for all data items with request rate
the desired
percentile. For ESRATE, the
confidence interval is given by-
^
Pf Z
^ 2f
Nk
Page
44
percentile
1 1
k = ( +1) ,
2
When
(2 k1)P f 1
for any data item
with
Pf
. Based
on N, for these data item the minimum request sample size N in order to satisfy the
accuracy required is given by-
N=
3 Z 2
6 Z 2
=
k 2 ( 1 +1) 2
The minimum value of
2
f
^
Z
Nk
^ 2f
k2
4 k1
=
Nk
2
k2
4 k1
for any given
k , then-
2
1
+1
2
3 +1
( )
( )
Therefore ESRATE scheme will meet the accuracy requirement for any data item
with
Pf >
as long as
2
1
+1
2
3 +1
( )
( )
4.5.1 Analysis
The analysis of ESRATE is significantly more complicated than RATE, CATE and Fast
Top-K. This is because that the different comparison in ESRATE is not independent.
Therefore we need to account for the covariance between different comparisons in order
to accurately compute the variance of estimation of proportions. In this section we
Page
45
concentrate on data item and assume that the Stateless Predecessor Table (SPT)
maintains all k-requests to the system. Therefore when there are total N requests to the
system. We would have made
Nk
comparisons. We make less than k comparisons for
the first k request arrival since k is small compared to N; we ignore this in the rest of
the analysis.
We assume that requests to system are approached as iid (Independent
and Identically Distributed random variable) [8] where the probability that request
belongs to data item
is given by
Pf . We label the request arrival 1 to N based on
request arrivals sequence. Let-
Cijs ( f )= 1 if i, j , s f
0, ot h erw ise
Let
M (N , f )
denote the number of requests for data item
after N requests.
Thereforei1
M ( N , f )=
i N j =ik
Cijs ( f )
4.5.2 Correlations among the comparisons

Before study the correlation structure of the comparisons, we state the following
elementary result first-
Lemma 2: Let
Cijs (f )
be defined above. Then-
E [ Cijs (f ) ]=P f
Page
46
VAR [ C ijs (f ) ]=P f ( 1Pf )
Proof: this result follows directly from the assumption that request are independent and
probability that an arrival belongs to data item
is
Pf . But in ESRATE, the
comparisons are not always independent of each other. Let use the comparison
Cijs (f )
and
Cimr (f )
P [C ijs ( f )=1 ]=P3f
where
due
to
P [ {Cimr (f )=1|Cijs (f )=1 } ]=P2f ,(r s)

request
( i j m) (r s)
as an example. Note that
independency
of
requests.
But
because the condition is already implies that
i f . In general for any pair of comparison
Cijs (f )
and
Cimr (f )
are
independent if and only if all the indices are distinct. If any two of the indices are
identical, then the comparisons are dependent. For example,
Cijs (f )
and
Cimr (f )
are dependent. The next result gives the correlation between the random variables
Cijs (f ) and Cimr (f ) .
Lemma 3: Consider
Cijs (f )
and
Cimr (f )
for
Then-
COV [ Cijs (f ) ,C imr (f ) ]=P 4f (1P2f )
Proof: Let
=COV ( Cijs ( f ) ,C imr ( f ) )
=E {C ijs ( f )E [ Cimr ( f ) ] }{C imr ( f )E [ Cimr ( f ) ] }
Page
47
ik j , m i1
and
rs .
=E {C ijs ( f ) Pf }{C imr ( f )Pf }

3
=E[ Cijs (f )C imr ( f )P3f C ijs ( f )P3f C imr (f )+ P6f ]
=E [ Cijs ( f ) C imr ( f ) ]2 P3f E [ C ijs ( f ) ]+ P 6f
=E [ Cijs ( f ) C imr ( f ) ]2 P6f +P6f
=P4f P 6f =P4f (1P2f )
Where the third equality follow from the fact that
last step follows from the fact that

request arrivals
probability
i, j , m
E [ Cijs ( f ) ]=E [ C imr (f ) ] . At
Cijs (f ) and Cimr ( f ) are both one if and only if
all belongs to the data item
and
P4f . In fact it is easy to show that the covariance is
r=s , which happen
P4f (1P2f )
for any
two comparisons that are correlated. We can now derive the Mean and Variance for the
f .
request of data item
4.5.3 Expectation and Variance of
Lemma 4: Let
M (N ,f )
M (N ,f )
denote the number of requests for data item
request arrivals to the system. Then3
E [ M ( N , f ) ] =Nk Pf
Page
48
after N
2(2 k1) P f
1+ Pf
VAR [ M ( N , f ) ]=Nk P4f ( 1P2f ) 1+
Proof: Note that-
[
i1
E [ M ( N , f ) ] =E
i N j=ik
C ijs ( f )
E [ M ( N , f ) ] =Nk Pf
To simplify the notation we assume that we index the comparisons using a single index
I m (f )
where
is set to one if comparison m result in a requests for data item
f . The variance can be computed as followsVAR [ M ( N , f ) ]=E [ M 2( N , f ) ]{E [M (N , f )]}2
E [ M ( N , f ) ] =Nk P3f
Now,
E [ M (N , f ) ]=E
i=1
[
Nk
[
Nk
i=1
Nk
I ( f )+
2
is
i=1 i=1 ;1 j Nk ; j i
I is ( f ) I js ( f )
{I 2i ( f ) } =Nk P4f
Now,
i=1;1 j Nk ; j i
I is ( f ) I js ( f ) =E
COV [ I is ( f ) I js ( f ) ]
Nk
i=1
COV [ I is ( f ) I ls ( f ) ]
I is ( f ) I js ( f ) +
j ;1 j Nk ; j i
Page
l ;1 l Nk ;l i
49
I is ( f ) I ls ( f ) }
Now, Variance becomesNk
VAR [ M ( N , f ) ] =Nk P4f + [ 2 ( 2 k1 ) P5f +{Nk12(2 k1)}P6f ]( Nk P3f )2

i=1
VAR [ M ( N , f ) ] =Nk Pf ( 1P f ) [1+
Notice that
2(2 k 1) Pf
]
1+ P f
Nk P4f ( 1P2f ) is the variance of
M ( N , f ) when all request samples are
independent from each other, therefore the correlations among request samples in
ESRATE increase the variance of
M (N , f )
by factor of
2( 2 k1)P f
. Since we
1+ Pf
know the mean and variance of the number of requests are matched. We now use the
Central Limit Theorem to obtain a normal approximation for the number of request
match than use the result to estimate the proportion. The next theorem gives the
expression for estimator of the proportion along with its variance.
Lemma 5: Let
M (N , f ) represents the number of requests for data item
N request arrivals for ESRATE with
Nk
Where,
comparisons for each arrival. Then-
M (N , k )
Pf N [0, 2f ]
Nk
(1P2f )
2f =
1+
2 (2 k 1 ) P f
1+ P f
Page
50
after
Proof: Though the comparisons are not independent, the comparisons are a stationary
k2
dependent sequence with finite exception and variance. The following the central
limit theorem for dependent sequence [1]. We can show that for large N-
Nk
Where,
M (N , k )
P3f N [0, 2f ]
Nk
2f =P4f ( 1P 2f ) 1+
2 ( 2 k 1 ) P f
1+ P f
^
Pf
Therefore the point estimates for
}
of
Pf
is-
M (N ,k)
^
Pf =
Nk
Then variance of estimation of
(1 ^P2f )
^ 2f =
1+
^f
2 (2 k 1 ) P
^f
1+ P
Pf
We know this expression for variance of the estimator and derive upper bounds on its
value. This bound is derived in two regions. The first upper bound on the variance hold
in the entire
[0,1]
range and is a function of
variance is a constant independent of
and holds when the proportion is below the
threshold.
k . The second bound on the
Page
51
Lemma 6: Let
2f
(1 ^P2f )
2f =
1+
^f
2 (2 k 1 ) P
^f
1+ P
then-
k
4 k1
2f 0.33if P f <
1
2k 1
Proof: Set the derivative of the variance with respect to
Pf <
upper bound. When
2f
1
2 k1
Pf
to zero gives us the first
32 P f P2f
9
This variance takes on a maximum value of
1
3
when
Pf =0
. The above bounds on
the variance can now be used to compute the sample size and estimation accuracy of
ESRATE. Note the setting
2
f
^ =
2
f
^ =
(1+2 Pf 3 P2f )
9
(1+2 Pf 3 P2f )
3
k =1 in variance of estimation, gives-
1
3
4.5.4 Minimum Sample Size

Page
52
(2 k1)P f 1 , let
desired
be the desired estimation accuracy and
percentile. For any data item
with
Lemma 6 its variance takes on a maximum value of
the
(2 k1) P f 1 , since based on

1
3
, the minimum request
sample size N in order to satisfy the accuracy requirement is given by2
3Z
N= 2
k
So, compare with other fast estimation approaches that2
ESRATE
CATE
=N
Fast Topk
3 Z
k
1
RATE
N
0.46 k
Now compare in case of processing speed & number of k comparisons-
ESRATE> Fast Topk >CATE> RATE
4.5.5 Request Weight Coefficient

We have to provide dynamically decreasing weight of request through estimated
coefficient for oppressive users to handle the fact that it is consolidates to assign static
request weight accurately. This weight fixed according to request rate. It protects
innocent users by discriminating oppressive users on heavy workload in Critical State.
At processing dynamic request weight coefficient that reduces the weight assigned to
oppressive users. For every request the mechanism counts the amount of traffic as total
number of request and request rate with used memory. It uses this amount to affect the
weight given to the innocent users. A time unit is set to roughly the expected burst time
Page
53
ARR
where
ARR
is average request rate. Let
R(t)
be request rate measured
in the requests is particular requests counter value divided by total requests arrives
since entering the system in Critical State. Let
be the amount of offered traffic
during last sliding window. Request Weight coefficient,
W ( r )=
W (r ) is-
{()
r 0,
r0

, >
Where,
r0
=R ( t ) . N , Threshold or average request burst time.
, Original Fixed Request Weight (by default set as 1).
M (N , f )
t ,
, Previous Request Rate at time
Nk
, Reduction Factor, configure by system.
We require that,
means
is greater, smaller the weight where
M (N ,f )
is binomial function.
With help of using request weight coefficient, we have to balancing load on basis
of Least Workload Left and Maximal Workload Left with RR, WFQ or APFQ [26] like
scheduling algorithm to transmit request to their particular state processing with
following State-Rank.
Page
54
4.5.6 Speedup
Lemma 7: Given the accuracy requirement described above in Section-V and N be the
number of request sample in CSRT is required for ESRATE-
3 Z2
N CSRT , Matching = 2
k
(2)
Let us define the request rate amplification factor
[1] for the rate change of data
item before and after the Matching as-
Rafter
Rbefore
(3)
Lemma 8: Given the accuracy requirement described in Section-V.

the
number
N SPT , Filtering
well as
of
request
samples
required
for
N CSRT , Matching
ESRATE-CSRT(Matching)
be
and
be the number of request samples required for ESRATE-SPT(Filtering). As
N DRT , Draining
be the number of request samples required for ESRATE-DRT
(Draining).
N SPT , Filtering =
N SPT , Filtering =
As well
as
N Draining
Draining.Therefore we have
N CSRT , Matching
2
N CSRT , MatchingN Draining

3
(4)
(5)
be the number of request samples
required for
speedup of the detection (Matching, Filtering, and
Page
55
Draining) process even with
factor on rate amplification factor due to historical
information filtering.
4.5.7 Memory Requirement
Now we focus on the worst case expected memory requirement for ESRATE [5]. The
memory requirement for the estimation scheme is proportional to the size of this subset
of data items for which request rate and request weight coefficient are maintained. So,
we calculate the worst case expected memory requirement due to dependencies of
comparisons with given the request sample size
for an allocation of
Pf .
Lemma 9: Given the specified accuracy requirement described in Lemma-1 of ESRATE

[5], the maximum expected memory [7] is-
at Regular State , E [ L ( N ) ] 1.174
Z 2
(6)
Z 2 1 2
at Critical State , E [ L ( N ) ] 1.174
+ [ Z k ]
2 3
3
E[ L ( N )] 1.174
Proof: where,
[ ]
Z 2
2
3
1 2
[ Z k]
3
(7)
1 2
[ Z k ] . Follow the results of Lemma-7 from ESRATE [5]. From
3
theorem-8,9 [11] the objective function is maximized when
1
P i= i
n
Page
56
n , the maximum expected list length through objective
Therefore a fixed value of

function is-
n 1exp
( Nk
n ))
3
We now want to determine the value of
Assuming that
for which this function is maximized.
is continous, and differentiating this expression with respect to
n . We get-
1exp
( )(
Nk
3
n
If we set
3 Nk
+1 =0
3
n
Nk
=
n3
we can rewrite the expression as-
1e (3 +1 )=0
Solving for
n=
we get
=1.903 . Therefore,
Nk
3
=0.81 Nk
1.903
Setting this value of
in objective function, the maximum expected memory
requirement is given by-
(
E[ L( N )] 0.81 Nk [ 1e
3
Nk
3
3
(0.81 Nk )
)]
Page
57
E[ L( N )] 0.687 Nk
Corollary-1 at Local Web Server or Regular State: given the width of the confidence
interval
and the error probability
E[ L ( N )]=0.687
3 Z 2
k 2
, the expected memory-
2
Z 23
3 Z
3
E[ L ( N )]=1.174
=1.174 Where , = 2
( )
This expression is particularly for local web server. But after performing analytical
simulation & trace memory usage we found that the memory size at proxy server has
been increased which at least equivalent to
Error,
1 2
[ Z k ]
3
named as Memory-Prediction-
. Reason behind the name for Memory-Prediction-Error is that through it we
predict about memory usage at proxy server as well as perform estimation for actual
memory used.
Corollary-II at Proxy Server or Critical State: given the width of the confidence interval
and the error probability
E [ L ( N ) ]=0.687
3 Z 2
k
k+
, the expected memory-
1 2
[Z k ]
3
2
Z 23 1 2
3 Z
3
1 2
E [ L ( N ) ]=1.174
+ [ Z k ] =1.174 + [ Z k ] Where , = 2
3
3
( )
Page
58
Summary: Let
L(T )
comparison up to
represents the number of data items that have at least one

arrivals. By Poisson process proved by Aldous, we have-
f =1
f =1
E [ L ( N ) ]=E [ f ( N )] 1exp (P 2f Nk)
And
E [ L ( N ) ] 0.638 Nk
Here
1 2
[Z k ]
3
1.11 Z 1 2
[ Z k ]
is Memory-Prediction-Error, factor of
1
3
is minimum variance &
is the worst case memory requirement as factor used in
comparison. So
reduce complexity of memory achieved at request migration in cloud & required

accuracy improvement in flash memory also, the memory amplification factor must be
used. It will be add when request in Critical State otherwise memory will reduce when it
in Regular State.
Page
59
Chapter-5
Implementation
Dont leave true & hardness during your work because your work is to discover your
world and then with all your heart give yourself to it.
-God Buddha, a Miracle Aviator & Founder of Buddhism, India
Through the performance evaluation, we target to answer following three questions
arise from some technical misconception of Workload Factoring and traffic estimation
over network through incoming requests for web services. We would like to solve such
difficulties
What is economical advantage of application hosting servers based on Integrated
Cloud Computing Model?

What is benefit on Regular State & Critical State to apply Extreme State-Rank
based Workload Factoring Scheme?

Is ESRATE possible to decrease the estimation time with respect to modest
increase or decrease in memory according to dynamic nature of workload &
system state?
For all questions we rely on Observation Tracery to evaluate performance criteria as

economic tendency, performance, memory cost & predictions as well as minimized data
cache & replication overhead of Integrated Cloud Computing Model in a large scale
configuration.
5.1 Tri-Way Trajectory Simulation

We set up an application Tri-Way Trajectory Simulation which hosts Yahoo!
Video-like video streaming service. The Tri-Way Trajectory Simulation consists of three
Page
60
parts: a local data center platform, a cloud service platform and a server with ESRWF
simulator (which is also part of LDC but for understanding we show server & LDC
differently) set up at large cloud service provider such as Amazon AWS EC2 and S3
services. In the local data center, around 15 open source RTP/RTSP streaming servers
such as QuickTime, Helix Universal & DNA, Wowza Media, Darwin, pvServer, Feng and
VideoLAN are provisioned all the time, while the streaming server instances are
activated on demand. The ESRWF component was implemented as a load controller for
the stream workload. When request for a video clip comes into the Tri-Way Trajectory
Simulation as a HTTP request, the Apache web server parses the request and asks
ESRWF for which stream server (either a LDC server or a cloud server) the video clip
will be served; it then returns with a dynamic HTML page which automatically starts
the media player for video streaming at the user machine.
Figure 5.1: Proposed Integrated Cloud Computing Model and Tri-Way Trajectory Simulation overview
The load controller also contains the implementation of two well-known load
balancing algorithms: the Least Connections balancing algorithm for the local data
center and the Round-Robin balancing algorithm for the cloud service provider. We
developed a distributed workload generator based on RTSP (Real Time Streaming
Protocol) to generate real video streaming load.
Page
61
5.2 Observation Tracery

As observed analysis, dedicated runtime implementation of ESRWF scheme is
developed through Tri-Way Trajectory Simulations as implemented part of proposed
Integrated Cloud Computing Model described in Figure 5.1. Yes, it is hard to obtain
analytic results for the performance of ESRWF under a stochastic environment. To
evaluate the behavior of ESRWF and efficiency of ESRATE in such an environment we
use simulation results. We conduct the simulation tool using the CloudSim-2.1.1 &
GridSim-5.2 [27] with network simulation tool OPNET Modeler 14.0 [28] and for
transmitting the request use the base line for applying the algorithm APFQ. The
implementation collects the statistics for each data item at simulation time in CSRT,
sums up the offered traffic load and updates the request rate and request weight
coefficient.
Figure 5.2: Video stream workload evolution on Yahoo! Video Site for ESRWF (January-March, 2012)
Now for active tracery, we used the Yahoo! Video (India) [9], is 3 rd ranked online video
website just after YouTube and MetaCafe in terms of total number of video views,
uploading & downloading during January 2011, delivered totally 30.2 millions of video
streams to 17.56 million of unique Indian viewers [10]. It is 72% of the online
Page
62
population of India who take services of video streaming. We trace the Yahoo! Video site
for 2 months (from January 11 to March 12, 2012), and the data was collected in every
30 minutes to 1 hour. Due to large scale of Yahoo! Video site, we limited the data
collection to the first 10 pages of each category. Since each page contains 10 video
objects, each time the measurement collects dynamic workload information for 1230
video files in total. Throughout the whole collection period, we recorded 2,843 unique
videos which durations range from 2 to 6350 seconds and a total of 1,755,186 video
views. This can be translated into a daily video request rate of 29253. There is nearly
dissimilarity in statistics to calculate mean duration between sites.
TABLE-I
Customer viewed video stream quality
Applied Schemes
Number of Unique Video Requested
Random
2420
Choke
2000
RATE
560
CATE
91
Fast Top-K
36
ESRATE
23
Page
63
Figure 5.3: Customer viewed Video stream quality at both system states
5.3 Requests Concurrency at servers

Figure 5.3 shows the viewed stream quality from stream resources of Regular State
(LDC Server) and stream resources of Critical State (Cloud Server) in the particular
Workload Factoring experiment. For informal comparison, both state servers have the
typical configuration & proficiency. We observed that the concurrent connections went
up; the Critical State server delivered more reliable streaming proficiency than the
Regular State server. It could corroborate 175 concurrent stream connections whereas
the Regular State server could only corroborate around 87 average concurrent stream
connections to keep the Customer-sided quality steadily at above 90 (see Figure 5.4). In
the Tri-Way Trajectory Simulation we set the Regular State server capacity at 90
concurrent connections and that for Critical State servers at 170. We observed that the
ratio of the max load to the avg load
()
was reduced to 1.94 in the Regular State
and streaming proficiency at Critical State is
=95
State.
TABLE-II
Page
64
more than that in the Regular
Customer viewed Concurrent Request stream quality at both state servers
Concurrent
Customer Viewed Quality at
Request Stream
Critical State Server
Regular State Server
Drain Resolver
12
12
12
16
16
16
20
20
20
24
24
24
28
28
28
32
32
32
36
36
36
10
40
40
40
11
44
44
44
12
48
48
48
13
52
52
52
14
56
56
56
15
60
60
60
16
64
64
64
17
68
68
68
18
72
72
72
19
76
76
76
20
80
80
80
Connections
Page
65
21
84
84
84
22
88
87
87
23
92
86
87
24
96
85
87
25
90
84
87
26
91
83
87
27
92
82
87
28
93
81
87
29
94
80
87
30
95
87
87
31
96
86
87
32
90
85
87
33
91
84
87
34
92
83
87
35
93
82
87
36
94
81
87
37
95
80
87
38
96
87
87
39
90
86
87
40
92
85
87
41
94
84
87
42
96
83
87
43
98
82
87
44
100
81
96
45
102
80
98
Page
66
46
90
87
87
47
93
85
87
48
96
83
87
49
99
81
87
50
97
79
87
51
95
77
87
52
93
84
87
53
90
83
87
54
94
82
87
55
98
81
87
56
94
80
87
57
98
87
87
58
102
86
93
59
98
85
87
60
90
84
87
61
91
83
87
62
92
82
87
63
94
81
87
64
95
86
87
65
96
84
87
66
97
82
87
67
90
80
87
68
92
87
87
69
94
85
87
70
96
83
87
Page
67
71
98
81
87
72
100
79
95
73
102
77
95
74
90
75
87
75
92
73
87
76
94
71
87
77
96
71
87
78
98
84
87
79
100
83
95
80
92
82
87
81
90
81
87
82
92
80
87
83
94
79
87
84
96
78
87
85
98
85
87
86
90
75
87
87
95
74
87
88
96
80
87
89
97
80
87
90
98
70
87
91
99
70
87
92
90
70
87
93
101
85
90
94
102
83
92
95
92
81
87
Page
68
96
93
79
87
97
94
77
87
98
90
77
87
99
96
75
87
100
97
75
87
101
90
85
87
102
92
75
87
103
94
65
87
104
90
55
87
105
98
55
87
106
100
57
92
107
92
58
87
108
94
60
87
109
96
70
87
110
89
80
87
111
100
80
95
112
102
75
95
113
91
65
87
114
93
60
87
115
95
55
87
116
90
50
87
117
101
50
92
118
101
52
92
119
95
54
87
120
96
75
87
Page
69
121
97
80
87
122
97
80
87
123
99
45
87
124
100
35
95
125
91
25
87
126
91
25
87
127
95
21
87
128
90
21
87
129
101
75
92
130
101
75
95
131
102
50
98
132
97
15
87
133
98
15
87
134
87
15
87
135
100
20
92
136
101
25
95
137
102
30
98
138
90
40
87
139
92
55
87
140
96
65
87
141
96
75
87
142
98
75
87
143
100
78
95
144
102
78
98
145
92
60
87
Page
70
146
92
50
87
147
94
40
87
148
96
30
87
149
98
20
87
150
100
10
89
151
102
91
152
92
87
153
92
87
154
94
87
155
96
57
87
156
101
61
95
157
102
65
98
158
91
69
87
159
91
72
87
160
102
72
89
161
102
60
91
162
104
45
93
163
104
30
95
164
89
15
87
165
93
87
166
99
87
167
102
97
168
105
20
98
169
108
25
99
170
92
30
87
Page
71
171
94
66
87
172
98
68
87
173
102
70
99
Figure 5.4: Customer viewed concurrent request stream quality at both system states
5.4 Memory Cost & Predictions

As comparison, unlike RATE [6], CATE [7] and Fast Top-k [1]; ESRATE is most
accurate and fastest uses less memory to follow the request migration policy of Extreme
State-Rank based Workload Factoring Scheme but addition of memory prediction
estimation
at certain level due to system state through which scheme is more
complicated and have desired high accuracy with sufficient ample stock of memory at
Critical State. So, such dynamic tenancy of memory is effective runtime example of
creation, storing & processing of memory block according to provisioning of incoming
request. As Figure 5.7, we also note that ESRATE use the typical amount of memory by
the time they capture all medium-sized requests, which signifies that they need less
memory size for estimation accuracy at both states. As a prior think is that to see
Figure 5.10, this technique leads only best resource planning & management appeal
Page
72
with comparison of IWF (Intelligent Workload Factoring) [1], which perform better in
each case and may improve the quality of the IWF.
TABLE-III
Memory cost for ESRWF
Number of Requests in memory at
Total Number of Request Arrived
Critical State
Regular State
0.00E+00
1.00E+00
1.00E+00
1.00E+01
1.00E+02
10
10
1.00E+03
15
20
1.50E+04
20
30
2.00E+05
30
40
3.50E+05
40
50
4.00E+05
50
60
4.50E+05
60
70
5.00E+05
70
80
5.50E+05
80
90
6.00E+05
90
100
6.49E+05
100
200
6.78E+05
200
300
6.98E+05
300
400
7.17E+05
400
410
Page
73
7.73E+05
500
420
7.91E+05
600
430
8.50E+05
700
440
9.50E+05
800
450
1.00E+06
850
460
1.25E+06
900
470
1.65E+06
920
480
1.88E+06
930
490
2.00E+06
940
500
2.16E+06
950
510
2.56E+06
960
520
2.91E+06
970
530
3.13E+06
980
540
4.22E+06
990
550
5.13E+06
1000
560
6.31E+06
1010
570
7.45E+06
1020
580
8.65E+06
1030
590
9.78E+06
1040
600
1.00E+07
1050
610
1.11E+07
1060
620
1.24E+07
1070
630
1.30E+07
1080
640
1.40E+07
1090
650
1.55E+07
1100
660
Page
74
1.72E+07
1110
670
1.90E+07
1120
680
5.5 Minimizing Data Cache and Replication Overhead

Linear approach of data accessing, prediction, servicing, caching, requesting and
drain or dispatching through which our Workload Factoring scheme reduces data
cache and replication overhead. It brings multiple and multitask benefits on authors
model & have effective job & performance in Critical State. Yes, In Critical State data
storage hierarchies based on ILMs designed as a data cache. And disengage from
Regular State through short extreme activated application of data accessing. Taking
prediction of incoming requests for particular data item & their related request rate
proportion through which system servicing with speeding up the dynamic provisioning
process at runtime server (Local Data Centre or Server in Cloud); cached data can
enlarge the capacity for particular runtime server for momentary relaxation. After
applying drain or dispatching, only small sized request of data item will be requested.
Also, apply WFQ-APFQ [26], Least Connections or RR type load balancing schemes [17]
as well as make use of and derive benefit from content locality which based on High
Workload based Dispatching.
Page
75
Figure 5.5: Workload factoring performance: Memory Cost in Critical State for ESRWF
Figure 5.6: Workload factoring performance: Memory Cost in Regular State for ESRWF
Figure 5.7: Workload factoring performance: Memory Cost at both State (Combined) for ESRWF
5.6 Allocated Bandwidth
Page
76
A fundamental problem of Integrated Cloud Computing Model is bandwidth

estimation: determining the bandwidth (bits/sec) required to carry traffic with a specific
bit rate (bits/sec) offered to an Internet link and satisfy quality-of-service requirements.
The traffic is requests of varying sizes arriving for transmission on the link. Requests
can queue up and are dropped if the queue size (bits) is bigger than the size of the
buffer (bits) for the queue. A link has a bandwidth (bits/sec), the rate at which the bits
of requests are put on the link. Over an interval of time during which the traffic is
stationary, the requests arrive for transmission at a certain rate, the traffic bit rate
(bits/sec), which is defined, formally, to be the mean of the sample sizes (bits) divided by
the mean request inter-arrival time (sec), but this is approximately the mean number of
arriving bits over the interval divided by the interval length (sec). Over the interval there
is a mean simultaneous active connection load, the mean number of source-destination
pairs of hosts actively sending requests over the link. The utilization of the link is the
traffic bit rate divided by the bandwidth; it measures the traffic rate relative to the
capacity of the link. This article presents results on a fundamental problem of
engineering the Internet. Our approach to solving the bandwidth estimation problem is
to use queuing theory and queuing simulations to build a model for the QoS
bandwidth. The traffic inputs are live streams from measurements of live links and
synthetic streams from statistical models for traffic streams as shown in Figure 5.8 &
5.9.
TABLE-IV
Request serve through ESRWF with allocated bandwidth at both system states
Number
of
Bandwidt
h
Allocated
(kbps)
250.00
Total
Number of General Requests
Number of Oppressive Requests
(%)
(%)
No. of
Reques
General
ESRWF
General
ESRWF
Arrived
Schemes
Schemes
Schemes
Schemes
0.81
0.52
0.63
0.54
0.52
Page
77
260.00
0.87
0.54
0.72
0.56
0.54
270.00
0.88
0.54
0.70
0.59
0.56
280.00
0.90
0.57
0.73
0.61
0.58
290.00
0.96
0.56
0.79
0.63
0.59
300.00
0.95
0.60
0.78
0.65
0.61
310.00
0.97
0.61
0.81
0.67
0.64
320.00
0.93
0.60
0.78
0.70
0.67
330.00
0.91
0.62
0.85
0.73
0.68
340.00
0.95
0.66
0.85
0.72
0.71
350.00
0.96
0.65
0.91
0.75
0.73
360.00
0.97
0.68
0.85
0.79
0.75
370.00
0.98
0.69
0.88
0.80
0.79
380.00
0.96
0.72
0.93
0.84
0.81
390.00
0.98
0.73
0.92
0.86
0.82
400.00
0.98
0.71
0.93
0.88
0.85
410.00
0.96
0.75
0.93
0.89
0.84
420.00
0.99
0.78
0.95
0.90
0.87
430.00
1.00
0.80
0.96
0.90
0.87
440.00
0.99
0.81
0.92
0.91
0.90
450.00
1.00
0.86
0.97
0.93
0.90
460.00
1.00
0.84
0.95
0.93
0.91
470.00
0.99
0.83
0.97
0.93
0.90
480.00
0.99
0.86
0.97
0.95
0.92
490.00
0.98
0.93
0.99
0.97
0.92
500.00
1.00
0.89
0.99
0.96
0.93
Page
78
510.00
1.00
0.84
1.00
0.96
0.94
520.00
0.99
0.91
0.99
0.97
0.94
530.00
1.00
0.91
1.00
0.96
0.95
540.00
0.99
0.92
1.00
0.98
0.95
550.00
1.00
0.90
1.00
0.97
0.97
560.00
1.00
0.92
1.00
0.98
0.96
570.00
1.00
0.95
1.00
0.99
0.97
580.00
1.00
0.93
1.00
0.98
0.96
590.00
1.00
0.98
1.00
0.99
0.98
600.00
1.00
0.96
0.98
0.99
0.97
610.00
1.00
0.95
1.00
0.99
0.98
620.00
0.99
0.97
1.00
0.99
0.99
630.00
1.00
0.96
1.00
0.99
0.98
640.00
1.00
0.96
1.00
0.99
0.99
Figure 5.8: Workload factoring performance: Allocated Bandwidth (Regular Requests) for ESRWF
Page
79
Figure 5.9: Workload factoring performance: Allocated Bandwidth (Oppressive Requests) for ESRWF
5.7 Compare with previous schemes

For fair comparison, we see when the concurrent connections went up; the
critical state server of ESRWF delivered more reliable streaming quality than the
trespassing zone server of IWF. It could support up to 170 concurrent stream
connections while keeping the client-side quality at above 90, whereas IWF could only
support around 160 concurrent stream connections to keep the client-sided quality
steadily at above 80. So, we compare both schemes on the basis of workload strategies.
TABLE-V
Comparison between IWF & ESRWF basis of workload strategy
Workload
Strategy
Number of General Request (%)
Number of Oppressive Request (%)
IWF Scheme
ESRWF Scheme
IWF Scheme
ESRWF Scheme
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Page
80
1.00
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
1.00
0.99
1.00
1.00
1.00
0.98
1.00
1.00
1.00
1.00
10
1.00
1.00
1.00
0.99
11
1.00
1.00
1.00
0.98
12
0.99
1.00
0.99
0.99
13
0.99
1.00
1.00
0.97
14
0.97
1.00
0.99
0.95
15
0.96
1.00
0.98
0.94
16
0.94
1.00
0.98
0.92
17
0.91
1.00
0.96
0.90
18
0.89
1.00
0.97
0.87
19
0.87
1.00
0.95
0.82
20
0.85
1.00
0.93
0.78
21
0.81
1.00
0.93
0.77
22
0.79
1.00
0.90
0.72
23
0.78
1.00
0.89
0.71
24
0.76
1.00
0.88
0.67
25
0.74
1.00
0.86
0.65
26
0.72
1.00
0.84
0.63
27
0.71
1.00
0.83
0.61
28
0.69
1.00
0.82
0.60
Page
81
29
0.67
0.99
0.80
0.58
30
0.66
1.00
0.79
0.57
31
0.64
0.99
0.77
0.55
32
0.63
1.00
0.76
0.53
33
0.62
0.98
0.75
0.53
34
0.61
0.96
0.74
0.52
35
0.60
0.95
0.73
0.51
36
0.59
0.92
0.72
0.50
37
0.57
0.89
0.70
0.50
38
0.56
0.88
0.70
0.50
39
0.56
0.86
0.69
0.49
40
0.54
0.85
0.68
0.48
41
0.53
0.84
0.67
0.48
42
0.53
0.82
0.66
0.47
43
0.51
0.81
0.65
0.47
44
0.50
0.80
0.65
0.45
45
0.50
0.79
0.64
0.45
46
0.49
0.77
0.63
0.45
47
0.48
0.74
0.62
0.45
48
0.47
0.74
0.62
0.44
49
0.47
0.72
0.61
0.44
50
0.46
0.72
0.60
0.43
Page
82
Figure 5.10: Requests serve through ESRWF at Frequency of Workload Strategy
Figure 5.11: Workload Handling Strategy managed by IWF & ESRWF (Adopting of Requests)
Page
83
Figure 5.12: Formulative Descriptions for Workload Factoring Scheme through ESRATE
5.8 Economical Tendency

We give an ESRWF based application hosting solutions to host the measured
Yahoo! Video stream load. In this solution, a local data center is over provisioned over
regular workload (96.4%), and cloud based platform such as Amazon EC2 [12] is rented
on high demand for critical workload (3.6%). It offers an economical solution for green
Cloud Computing environment with 65-server oriented tiny local data center with 15server oriented Drain Resolver and an annual rent for managing critical workload is
approximately $42.56K only. As formality, we only include the server cost only though
so many factors which influence the server cost had been neglected. We measure the
application hosting solution for Yahoo! Video stream loads according to Methodology of
[1]. Figure 5.2 shows the unique video requested by customer were viewed & served only
23 out of 5000 video streams in Critical State. And Figure 5.12 shows formulative
description of particular requests which have been served. Its proved that ESRWF
scheme for Integrated Cloud Computing Model has excellent performance over network
resource proficiency enhanced with distributed network as well as parallel also.
TABLE-VI
Page
84
Performance criteria of ESRWF
Hosting Platform Issues
Solutions through ESRWF
Unique Video Requested
23 unique video clips
Equipment Required
65-server LDC + 15-Drain Resolver
Annual Costs
USD 42.56K approx.
Requests Concurrency
90 at Regular & 175 at Critical State
Resource Proficiency
=95
Ratio of maximum load to average load
=1.94
Data Cache & Replication Overhead
Minimized up to three orders of magnitude
Figure 5.13: Fair Comparison between IWF & ESRWF (for RAM, CPU Speed, Speed up, Memory Used)
Page
85
Chapter- 6
Conclusion & Future Works
In this life, we cannot do great things. We can only do small things with great inspirable
technologies or inventions with motivation or love.
-Arstole, Great Greek Philosopher, Greece
In this paper, we present the design of an Extreme State-Rank based
Integrated Cloud Computing Model. For the future work, extending the Integrated Cloud
Computing Model to stateful & provision based applications such as n-tier web services
is a natural and challenging step. Many new problems arise such as session
maintenance, service time estimation, and data consistency due to data dispatching in
different states to process. We are working on a fast data on demand service and
integrating the dynamic web service scaling approach in our system; there are many
concepts missed for discussion such as load balancing schemes in states, data
replication & consistency management in the Critical State,0 security management for a
Integrated cloud platform, and more. In this paper we have also developed an Extreme
State-Rank bAsed Traffic Estimation scheme called ESRATE that uses Matching &
Counting between different arriving requests to estimate the request rate. We have
shown through both theoretical analyses of Observation Tracery and Tri-Way Trajectory
Simulations that ESRATE is fast and memory-efficient. It estimates small data items
very quickly and accurately, and still gives good estimation accuracy for large data
items. Dependence within request arrivals may affect the estimation accuracy. Although
the buffer-based approach can alleviate the problem, a better approach that can further
minimize the impact of dependence without additional over-head may be more
preferable. We attempt to address this issue as part of future work also.
Page
86
We focus on the Workload Factoring component in this paper as it is a unique

functionality requirement by the Resource Management Architecture for Integrated Cloud
Computing Model. With the proposed Workload Factoring technology, the Extreme StateRank based Cloud Computing Model allows enterprise IT systems to adopt an
Integrated Cloud Computing Model where a dedicated resource platform runs for
hosting application in regular loads, and a separate and shared resource platform
serves in critical high load. Given the elastic natural environmental of the green cloud
infrastructure, it develops a situation where cloud resources are used as supplementary
extension of existing unable infrastructure to handle high workloads. It's not an all or
nothing decision; companies can effort into the cloud without leaving permanently
established infrastructure and applications.
Page
87
References
[1]
H. Zhang, G. Jiang, K. Yoshihira, H. Chen, and A. Saxena, Intelligent workload factoring

for a hybrid Cloud Computing model, NEC Labs America Technical Report 2009-L036,
Feb 2009.
[2]
M. Arregoces and M. Portolani, Data Center Fundamentals, Cisco Press, 2003.
[3]
Wikipedia- ILM, http://www.wikipedia.com/.
[4]
G. Karypis and V. Kumar, Multilevel k-way hypergraph partitioning, in DAC 99:

Proceedings of the 36th ACM/IEEE conference on Design automation. New York, NY,
USA: ACM, 1999, pp. 343-348.
[5]
Snehil Sharma, Abhishek Mathur and Shailendra Shrivastava, ESRATE: A Fast,

Memory-Efficient Request Rate Estimation by State-Rank based Scheme for Cloud
Computing, Unpublished.
[6]
M. Kodialam, T. V. Lakshman, and S. Mohanty, "Runs bAsed Traffic Estimator (RATE): A

simple, Memory Efficient Scheme for Per-Flow Rate Estimation", Proceedings of
INFOCOM'2004.
[7]
F. Hao, M. S. Kodialam, T. V. Lakshman, and H. Zhang, Fast, memory-efficient traffic

estimation by coincidence counting. In INFOCOM, 2005, pp. 2080-2090.
[8]
Wikipedia- IID, http://www.wikipedia.com/.
[9]
Yahoo! Video, http://video.yahoo.com/.
[10]
ComScore Video Metrix report: India Viewers Watched an Average of 3 Hours of Online
Video in July, Available:
http://www.comscore.com/Press_Events/Press_Release/2011/3/7_of_10_Indian_Web_U
sers_Watch_Online_Video_in_a_Month/.
[11]
Youtube, http://www.youtube.com/.
[12]
Amazon web services, http://aws.amazon.com/.
[13]
Google App Engine, http:code.google.com/appengine/.
[14]
Wikipedia- open source RTP/ RTSP Streaming Server, http://www.wikipedia.com/.
[15]
Darwin streaming server, Available:
Page
88
http://developer.apple.com/darwin/projects/streaming/.
[16]
Openrtsp, Available:
http://www.live555.com/openRTSP/.
[17]
Wikipedia- CRM,http://www.wikipedia.com/.
[18]
R. Pan, B. Prabhakar, and K. Psounis, Choke - a stateless active queue management

scheme for approximating fair bandwidth allocation, in INFOCOM 2000. Nineteenth
Annual Joint Conference of the IEEE Computer and Communications Societies.
Proceedings. IEEE, vol. 2, 2000, pp. 942.951 vol.2. Available:
http://dx.doi.org/10.1109/INFCOM.2000.832269
[19]
A. Kumar, M. Sung, J. xu, and J. Wang, Data streaming algorithms for efficient and
accurate estimation of flow distribution, in Proc. Of ACM SIGMETRICS, June 2004, to
appear.
[20]
M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. H. Katz, A. Konwinski, G. Lee, D. A.

Patterson, A. Rabkin, I. Stoica, and M. Zaharia, Above the clouds: A berkeley view of
cloud computing, EECS Department, University of California, Berkeley, Tech. Rep.
UCB/EECS-2009-28, Feb 2009. [Online]. Available:
http://www.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-28.html
[21]
Anthony T. Valte, Toby J. Valte, Robert Elsenpeter. Cloud Computing: A Practical

Approach. Tata McGraw-Hill Edition 2010, 6th Reprint.
[22]
Dr. Kumar Saurabh, Cloud Computing: Insights into New Era Infrastructure. WILEYINDIA, First Edition-2011.
[23]
A Hub & Spoke Model for Spatial Data Infrastructure. Available:

http://www.ec-gis.org/workshops/11ec-gis/presentations/17wal.... [Online].
[24]
Amazon, http://aws.amazon.com/cloudfront/.
[25]
Vmware cloud vservices, Available:

http://www.vmware.com/technology/virtual-datacenter-os/cloud-vservices/
Anat Bremler-Barr, Nil Halachmi, Hanoch Levy, Aggressiveness protective Fair Queueing
[26]
for Bursty Applications . Available:

http://www.cs.tau.ac.il/hanoch/Papers/BremlerHalachmiLevy2005.pdf.
[27]
CloudSim & GridSim API (Framework), Available:

http://www.cloudbus.org/.
[28]
OPNET Modeler 14.0, http://www.opnet.com/.
[29]
Oracle Technology Network, Cloud Computing Center, Available:

http://www.oracle.com/technology/tech/cloud/index.html, retrieved 14th May 2009
Page
89
[30]
Farber D., Oracles Ellison Nails Cloud Computing, in CNET Outside the Lines,
Available:
http://news.cnet.com/8301-13953_3-10052188-80.html, retrieved 14th May 2009
[31]
Boss G., Malladi P., Quan D., Legregni L., Hall H., IBM on Cloud Computing; High
Performance On-Demand Solutions, IBM, 8th Oct 2007
[32]
Armbrust M., Fox A. et al, Above the Clouds: A Berkeley View of Cloud Computing,
Available:
http://radlab.cs.berkeley.edu/; Feb 10, 2009
[33]
Twenty-One Experts Define Cloud Computing, Cloud Computing Journal, July 2008;
http://cloudcomputing.syscon.com/node/612375/print
[34]
Buyya R., Yeo C.S., Venugopal S., Market-Oriented Cloud Computing: Vision, Hype, and
Reality for Delivering IT Services as Computing Utilities, Keynote Paper, Proceedings of
the 10th IEEE International Conference on High Performance Computing and
Communications, Sept. 25-27, 2008, Dalian, China
[35]
Geelan J., Cloud Storage: How Can Enterprises Build Secure Private Clouds, Available:
http://cloudcomputing.sys-con.com/node/830646; Cloud Computing Journal; Feb 29,
2009
[36]
Duffield N., Lund C., and Thorup M., Charging from Sampled Network Usage,
SIGCOMM internet Workshop 2001.
[37]
Duffield N, and Grossglauser M., Trajectory Sampling for Direct Traffic Observation,
Proceedings of ACM SIGCOMM 2000.
[38]
Hao F., Kodialam M., and Lakshman T. V., "ACCEL-RATE: a faster mechanism for
memory efficient per-flow traffic estimation", proceedings of ACM SIGMETRICS 2004.
[39]
Fang W., and Peterson L., Inter-as Traffic Patterns and their Implications, Proceedings of
IEEE GLOBECOM 1999.
[40]
Feldmann A. et al., Deriving Traffic Demands for Operational IP Networks: Methodology

and Experience, Proceedings of ACM SIGCOMM 2000.
[41]
Feng W. et al., The Blue Queue Management Algorithms, IEEE/ACM Transactions on

Networking, Vol.10, Number 4, 2002.
[42]
Aldous D., Probability Approximations via the Poisson Clumping Heuristic, SpringerVerlag, 1987.
Page
90
[43]
Estan C. and Varghese, G., "New Directions in Traffic Measurement and Accounting",
Proceedings of ACM SIGCOMM 2002.
[44]
Ferguson T. S., "A Course in Large Sample Theory", Chapman and Hall, 1996.
[45] Estan C., Savage, S., and Varghese, G., Automatically Inferring Patterns of Resource
Consumption in Network Traffic", Proceedings of ACM SIGCOMM 2003.
Page
91

FFFFF

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

FFFFF

Diunggah oleh

Hak Cipta:

Format Tersedia

ESRWF: Extreme State-Rank based Workload Factoring for Integrated Cloud Computing Model

Samrat Ashok Technological Institute, Vidisha

By providing common definitions of the important terms;

problems inherent to it; and

1.2 Research Objective

complexity is hidden and resource multiplexing leads to proficiency. More computing

Samrat Ashok Technological Institute, Vidisha

Samrat Ashok Technological Institute, Vidisha

An efficient workload factoring philosophy is shaped for enabling technology

this workload factoring technology can enable reliable

1.3 Proposed Rationale

Opportunistically, achieves resource proficiency with reliable prediction on the

requirement of agile and responsive performance through cloud services.

Samrat Ashok Technological Institute, Vidisha

Integrated Cloud Computing Model

Distribution follows to categories the request according to states.

Samrat Ashok Technological Institute, Vidisha

Maintenance related to requests information privacy & security and their

Uses refer to state handling resource management services.

Disposition offers non-accepted requests draining or their neglecting services.

Exception related to agreement discontinuity or non-recurring issues outside

Required Goodness for Request Estimation Scheme

It should extremely estimate request rate to specified accuracy as well as proved

estimation is critical for fast detection of anomalous events.

or preferably amenable to cover each step of implementations.

For schemes concreteness, we have to aimed design goals of workload factoring

Reducing the higher workload and dynamic complexities in Regular State

Draining & Dispatching.

decomposition based on preliminary features of empirical threshold of system.

Samrat Ashok Technological Institute, Vidisha

1.4 Dissertation Outline

Samrat Ashok Technological Institute, Vidisha

2.1 Advent of Cloud Computing

Samrat Ashok Technological Institute, Vidisha

exposition of these capabilities to Consumers. Multiple new cloud domains and

Data centers trying to maintain high scalability and increase availability;

2.2 What is a Cloud?

Cloud computing is a emerging trends of Distributed Computing and

Samrat Ashok Technological Institute, Vidisha

Figure 2.2: Introduction to used Technology offered by Cloud Computing

2.3 Grid vs. Cloud Computing

parallel servers. [31] Wikipedia provides a vague, citation-less definition broadly

Figure 2.3: Grid Computing Working Scenario

Samrat Ashok Technological Institute, Vidisha

Figure 2.4: Cloud Computing Working Scenario

2.4 Cloud Computing: Definition

Samrat Ashok Technological Institute, Vidisha

the ability to pay-as-you-go as the necessary billing model, implying Utility

Computing as inherent to Cloud Computing and

UC Berkeley's definition on cloud computing is firm with regard to zero capital

Samrat Ashok Technological Institute, Vidisha

Cloud Provider SaaS Provider / Cloud User SaaS User

Infrastructure/Hardware as a Service and Platform as a Service, which is commonly

2.5 Essential Characteristics

Figure 2.5: Benefits through emerging trends of Cloud Computing

Samrat Ashok Technological Institute, Vidisha

On-demand self-service, a consumer can unilaterally provision computing

without requiring human interaction with each services provider.

client platforms (e.g., mobile phones, laptops, and PDAs).

processing, memory, network bandwidth, and virtual machines.

unlimited and can be purchased in any quantity at any time.

2.6 Service Models

Samrat Ashok Technological Institute, Vidisha