Anda di halaman 1dari 6



TIME Time University
Tunis, TUNISIA,,,

Abstract: In the MPLS architecture, there is a [I. MPLS

strong argument to virtualize network functions a. Definition
ahead of implementing SDN Because it is a
relatively new investment, the MPLS represents an MPLS is an ultimate solution in order to resolve the
opportunity to create a foundation architecture that traffic engineering problems (like speed, scalability,
will scale over. Moreover, there are two and Quality of Service (QoS) etc, encountered by
other closely linked in networking occurring in the today networks. MPLS allows routers at the edge of
same time NFV and SDN who can deliver benefits the provider network to incorporate labels into the
to the MPLS backbone and edge switching routers incoming packets from the customer side and then
which increase network component's forward traffic to the core routers based on labels
flexibility, scalability and reliability. rather than performing complex resource
consuming routing lookups. While the labels
Keywords: MPLS, OpenFlow, NetFlow, SDN, NFV, exchanged at the core are generally based on
E-LSR destination [P subnets, the labels are also dynamic
enough including more forwarding criteria like QoS
I. Introduction
and traffic engineering. [1]

Mu[tiple Protocols Labels Switching (MPLS) has

b. Architecture
evolved more than 15 years as a solution for [SPs.
At the beginning, it is used to perform traffic
The MPLS Provider's network contains 3 principle
engineering over IP networks and for offering its
based Layer 2 or Layer 3 enterprise Virtual Private
Network (VPN) services. The traffic engineering CE: A Customer Edge (CE) device. This is a router
and the profitable enterprise VPNs service are that connects to the customer network and to a
required for efficient use of the provider's network service provider.
architecture. [n the consequent of their limit, the PE: A Provider Edge (PE) device. This is a service
older technology like Asynchronous Transfer Mode provider piece of equipment that connects to a
(ATM) and Frame Relay will disappear, which customer and into the provider (P) network.
make MPLS the most recommended network for P: A Provider (P) device. This is a service provider
users. In the traditional provider's data plane, the
piece of equipment that exists entirely in the
MPLS represent an additional simple feature which
provider (P) network and only connects to other
can be integrated on the complex and expensive
service provider devices (not to customers) as
routers. In the control plane, the enhanced service
shown in figure 1.
increase the network cost and fragility which make
IPIMPLS too complicated. The Software Defined
Networks (SON) decouple the control plane and the
data forwarding plane defined respectively as
software-based and hardware-based. [n this case,
the forwarding decision is accorded to the control
plane, it program adequate path among the data
plane which take over the forwarding process. The
Network Function Virtualization (NFV) help on
reducing the capEX/opEX by the virtualization of
the networking components (routers, Switches,
firewall . . . ).
Figure 1 : MPLS basic architecture
The rest of the paper is structured as follows,
section II, introduction to the MPLS world, and [n addition, the PE and P routers are label switch

then section III, [V, V the definition of the Network routers. There are two types of label switch routers:

Virtualization. On section VI, VII the proposed

solution and then the conclusion. LSR: A Label Switch Router (LSR) is a
router/switch that is capable of forwarding packets

978-1-4673-6636-6/15/$31.00 ©2015 IEEE

The concept of network virtualization (NV) is
creating logical segmentation through a single
based on labels. The CE, or customer, devices are physical network. NV is achieved by installing
not LSRs and can handle regular unlabeled IP cluster of software to manage the services provided
packets. by the cloud computing. NV treats all network's
logical components and services as a single pool of
E-LSR: An Edge Label Switch Router (edge-LSR) resources that can be accessed without regard for
is a more specific term for the PE routers. An edge­ its physical location. The Network virtualization
LSR is an edge device that is also an LSR. For an decouple the control and the data planes as
MPLS network, this is the device that takes advocated by the SDN and the NFV.
unlabeled IP traffic and imposes, or in MPLS terms,
pushes an MPLS label and switches the traffic to IV. Software Defined Networks (SDN)
the next LSR. The edge-LSR also takes labeled
traffic and deposes, or in MPLS terms, pops the As defined by Open Network Foundation (ONF),
label and forwards it to the next hop. A PE device the SDN controller take the decision of switching or
is an Edge-LSR in MPLS-based networks. [2] engineering the flow, it use the OpenFlow control
protocol to interact with the SDN equipment.
The SDN stills evolving to improve the network's
functionality mentioning mobility, controlling and
programming capability to face the problems of
scalability and network performance, and certainly
the ability to adapt with multiple SDN controller or
SDN domains.

a. SDN architecture

Figure 3: SDN network architecture

The SDN network is composit of 4 essential

elements as shown in figure 3:

Figure 2: MPLS Plane architecture • Application Programming Interface

(API) Northbound
c. Meshing process
The API represents the interface between the
The meshing process take place as follows as controller's Software Module and the SDN
shown in figure 2: application on the network. It presents a set of data
1- Routers exchange information using model and functionality for the network
Interior Gateway Protocols (IGP) applications. The northbound API are open source­
2- Routers assign a local label to Routing based.
table addresses, and store it on the Label • The control plane
Information Base (LIB).
3- The locals labels are distributed among the The control plane presents a summary of the
others routers, and then stored on the network infrastructure which allow administrator to
Label Forwarding Information Base apply specific policies and protocols to the network
(LFIB) and Forwarding Information Base components. The controller take the decision of
(FIB). allowing, forwarding, choosing Path and QoS.
4- All routers build their LIB, LFIB and FIB Once the controller allow the flow, it programs the
tables. whole path for better forwarding conditions. . The
network operating system (NOX) controller is the
III. Network Virtualization most widely deployed controller.
• Southbound interface (SRI) data planes. The centralized manager define the
route and the data flow. It simplifies the managing
The SBI part defme the communications between of flow related to specific applications. It also
the controller platform and data plane devices, defines the priority setting for the network
including physical and virtual switches. The SSI application. Due to the distributed scale of network,
layer use various protocols like OpenFlow the centralized control require a high performance
forwarding protocols and NetFlow monitoring system and elastic infrastructure which make it
protocol. unwieldy and expensive.

• OpenFlow • Distributed SDN

OpenFlow was simply referred as "a way for The Distributed model contains centralized
researchers to run experimental protocols in the manager connected to Distributed controllers and
networks they use every day". The analogy used by data planes. The centralization of the manager
the authors in describing openflow is to think of interface makes system management appear as a
openflow as a general language or an instruction set whole of systems even when we are managing the
that lets one write a control program for the distributed network one by one. The distributed
network rather than having to rewrite all of code on model deployment is complex because of the access
each individual router. [3] and configuration difficulty, namely timing and
synchronization issues. Distributed model evolution
• NetFlow and scalability make it a recommended solution and
eliminates the Centralized model.
NetFlow is a networking protocol designed by
Cisco Systems for logging and recording the flow • Hybrid SDN

of traffic received and sent within a network.

NetFlow was designed in collaboration with From our perspective, there is value in both
Enterasys Switches. It provides network traffic centralized and distributed approaches, which is
statistics by collecting related data from enabled why we advocate a hybrid SDN model. In the
routers and switches. NetFlow may also be called hybrid model, you still have a centralized manager,
Cisco lOS NetFlow. but are distributing separate controller and data
planes. The hybrid SDN model leverages the
NetFlow is primarily designed for network benefits of the simple control of managing specific
administrators and managers to help them with data flows as in the centralized model with the
detailed information, statistics and overall network scalability and resiliency of the distributed model.
operation data. It is integrated within the Defming policies is key to any network, and
proprietary Cisco lOS installed on the supporting meeting an application's specific policy
routers and switches by default and works by requirements across a large network is very
registering all of the IP traffic that flows in and out complex. In a hybrid SDN, the ideal approach is to
of the network through these devices. have tightly bound policies in respect to certain
applications for more control, but loosely coupled
The information recorded by NetFlow is used for a networks for greater agility and flexibility in the
variety of network management and maintenance management of those networks. The hybrid
tasks such as traffic logging, usage statistics approach allows for more manageable policy
monitoring, and anomaly detection, which it definition in the sense that you can start small, or
accomplishes through network capacity planning more locally, and then as you expand in size, apply
and monitoring, and by monitoring unusual traffic those policies among more networking
trends. [6] components. For example, voice traffic may have a
more general global policy, however, if you want
• Data plane another application's traffic to take priority (such as
video traffic during a webinar broadcast) then you
It represents the physical equipment. SON data can defme the policy to one or more sets of network
plane provides high speed and more scalability paths without addressing it across the entire
using programmable equipment. The data plane network. Security prioritizations have the same
hardware asks the decision from the controller needs as general policies. You may want global
however sometimes it can decide by itself. security requirements to be managed in a
distributed fashion, but specific local security
b. SDN implementation models policies to be managed more centrally. It is much
more likely that network automation (which
translates into lower costs) will happen at the
The SDN network can be implemented as follow: controller level which is more tightly bound to
geography, while simpler global policies and
• Centralized SON security configuration will occur at the centralized
management level. [4]
The centralized model contains a centralized
manager working with a controller and a distributed
V. Network Function Virtualization committee group
(NFV) Table 1 comparison between the SDN and NFV

a. Definition d. Relationship NFV with SON

The white paper on NFV defines it as a NFV is highly complementary to SDN, but not
consolidation of Network functions onto industry­ dependent on it (or vice-versa). it can be
standard servers, switches and storage hardware implemented without a SDN being required,
located in DatalDistribution centers - an optimized although the two concepts and solutions can be
data plane under virtualization. NFV allows combined and potentially greater value accrued.
administrators to replace physical network devices NFV goals can be achieved using non-SON
(traditional) with software that is running on mechanisms, relying on the techniques currently in
commodity servers. This software realizes the use in many datacenters. But approaches relying on
'network functions' that were previously provided the separation of the control and data forwarding
by the dedicated hardware (network devices) [3] as planes as proposed by SDN can enhance
shown the below : performance, simplity compatibility with existing
deployments, and facilitate operation and
maintenance procedures.
NFV is able to support SON by providing the
infrastructure upon which the SON software can be
run. Furthermore, it aligns closely with the SDN
objectives to use commodity servers and switches.
We intend to work closely with organizations
progressing work on SDN such as the ONF whose
work we will specifically take into account. [7]

Figure 4: NFV architecture VI. The implementation of the SON and

NFV in the MPLS architecture
b. Benefits of the NFV

a. The general idea

The benefits of the NFV are:

The solution is to replace the E-LSR in the MPLS

• Flexibility: it helps operator on quick architecture with a NFV architecture containing a
service deployment with high flexibility. virtual router and virtual OpenFlow SDN controller
• Cost: it reduces the CapEX and the opEX which is based on information collected by the
cost for the network deployment. NetFlow SON Monitor.
• Scalability: it helps operator to scale their
network across datacenters servers to adapt b. OpenStack
to users changing needs and providing
services OpenStack is an open and scalable cloud operating
• Security: the security represents the major system for building public and private clouds. [t
challenges to provide network service provides both large and small organizations an
manage customer's secured virtual space. alternative to closed cloud environments, reducing
the risk of lock-in associated with proprietary
c. Comparison between NFV with platforms. It allows users to leverage large pools of
SDN on-demand, self-managed compute, storage and
networking resources to build efficient, automated
SDN NFV private and public cloud infrastructures managed
Motivation -Decoupling of -Abstraction of over a dashboard or via the OpenStack API.
control and data network
planes. function from
-Providing dedicated
centralized hardware
controller and appliances to
network COTS servers.
Network datacenters Service provider
location network c. Open Contrail
Network devices Servers and Servers and
switches switches The OpenContrail System consists of two main
components: the OpenContrail Controller and the
applications Cloud Firewalls,
OpenContrail vRouter. The OpenContrail
orchestration and gateways,
Controller is a logically centralized but physically
networking content delivery
distributed Software Defined Networking (SDN)

protocols OpenFlow N/A controller that is responsible for providing the

Standardization ONF ETSI NFV
management, control, and analytics functions of the
virtualized network.
The OpenContrail vRouter is a forwarding plane (of
a distributed router) that runs in the hypervisor of a
virtualized server. It extends the network from the
physical routers and switches in a data center into a
virtual overlay network hosted in the virtualized
servers. [8]

Figure 5: The architecture of the VR!VSC

d. Logical Components

In this architecture we will use 3 essential elements:

• Virtual Router (VR): the Component is a
Virtual Machine (VM) working as a P E
router which will treat customer request.
• Virtual SON Controller (VSC): it uses
OpenFlow protocol to make the
forwarding decision, the objective of using
Virtualized SDN Controller is decreasing
Figure 6: the architecture of the VSM/VSC
of latency and securing the communication
between the Edge Router and the SDN The Proposed architecture of the MPLS will be as
controller. The VSC is generally used to
shown in figure 7:
reduce latency and decrease bandwidth
• Virtual SON Monitor (VSM): it uses
NetFlow protocol to collect and monitor
the status of network components.

e. The proposed architecture

Normally, the change will appear on the network

Edge because it the most important part of network.

We add 2 new physical components as shown in the

two figures 5, 6:
• VRIVSC: it contains a VSC who makes
the forwarding decision and the VR which
will replace the E-LSR router and work as
the interface between the data plane and Figure 7: the new proposed architecture
the control plane. They work on the same
hypervisor so they will communicate on f. The VRlVSC functionality
the virtual network via API which will
reduce the latency. The VSC will treat the As we said, the VRlVSC is an E-LSR with a
packet received by the VR and then choose separated control and data plane. This separation
the forwarding decision based on provider will makes the decision better then who's made by
link information. the traditional router. The SDN control plane
increase the quality of the path using more criteria
• VSMIVSC: By dint of it composition, the like (the backbone's routers status, its unused
VSMIVSC will perform the centralized bandwidth, its link status, its capability, user's
manager. It collects the information about requirement . . . ). To improve the direct assignment,
all the routers and the LSR and the virtual resources configuration and satisfy the
backbone and then store it on the database. customer demands, the VSC most coordinate and
To make a robust information database, orchestrate a dynamic network configuration which
the user's configuration will be stored and will help on the implementation of the requirements
ordered on it and then it will present the at specific levels and takes the responsibility of
VRlVSC's information database. coordinating the delivery of the demands of a
customer request into the service that embodies the
requirement of that request.

g. Distributed VSC
• Reactive: just program the switches
The Distribution of the SDN Controller makes accessible by the current controller.
network more scalable when adding a new • Proactive: using inter-controller
controller to the cluster and reliable when saving communication to program all switch end
the switches who's connected to a failed controller. to end. [5]
The challenge emerging with this model are The reactive approach can used if all controller are
synchronization, the coordination and cooperation available and are full synchronized, if one
between the VSC. controller fail, it data plane will be blocked, which
leverage on network availability.
h. The VSMIVSC functionality The proactive approach is recommended because
the controller work together to program the
The principle function of the VSMIVSC is helping switches. In this case, if a controller fail the others
the VRlVSC with interpreting information about will replace it transparently.
the infrastructure components. It collect a database
of real-time monitoring information and send it as VIII. Conclusion
an update to all VRlVSC to be considered when
making decision. The network domain is always broaden, and with
It's recommended to use a centralized VSMIVSC, SDN and NFV capabilities, it will quickly develop.
to increase the reliability of information sources, The idea is to involve the newer technology with
resolve the Split brain, coordination, and older Concepts to makes a network more secure and
cooperation and synchronization problems. To to improve its availability.
minimize update traffic among the network, an
update timing most be configured, the timing
depend of the network latency and used bandwidth. REFERENCE
To cancel the update redundancy, the VSMIVSC [1] Analysis of traffic engineering parameters while using Multi­
will use revision number will be sent to distinguish protocol label switching (MPLS) and traditional IP
Networks written by Faiz Ahmed & Dr. Irfan Zafar
the older update.
Connecting the VSMIVSC directly to the Backbone
[2] Cisco systems, "CCIP MPLS Study Guide" James Reagn.
ameliorate the security level by using the
programmable firewall functionality on the SDN
[3] "SON and its use-cases-NV and NFV, a state-of-the-art­
data plane.
survey, written by Sridhar K. N. Rao

VII. The VSMIVSC communication layer

[4] http://www.virtualnetwork.comiblogl2013/08/centralized-vs­
a. Controller Service Discovery (CSD) todays-networksl 18 march 2015

In distributed controller case, when initiating the [5] "Controlling a Software-Defined Network via Distributed
network, the controller assign each switch a virtual Controllers" written by Volkan Yazici, Oguz Sunay, Ali
Qzer Ercan
IP addresses to use when communicate with its
controller. When the switch tries to communicate
[6] 22
with the controller, it get redirected to the March 2015
corresponding one.
In the edge of our proposed architecture, controller [7] Network Functions virtualisation - Intoductory White Paper
is attached to the router which reduce latency and
secure the communication. When the attached [8] http://www.opencontrai1.orgiopencontrail-architecture-
controller fail, the Router will redirected to other documentation/#section2 10 juin 2015
contro IIer.
The backbone architecture will be monitored by the
VSMIVSC, and it will be controlled by the
VRIVSC controller because the forwarding path
will depend of the packet treated there.

b. Inter-controller routing

The routing between inter-controlling has 2