A MULTI-OBJECTIVE LOAD

BALANCING
SYSTEM FOR CLOUD ENVIRONMENTS
Muhamamad Junaid Ali (CIIT/SP18-RCS-028)

1
 TABLE OF CONTENTS

• Introduction
• Related Work
• Multi-Objective Load Balancing System
• MO-LB System Model
• Particle Swarm Optimization
• Evaluation and Results
• Conclusion and Future Work
• References

2
 INTRODUCTION(1/3)

• Cloud Computing consists of three components:

1. Client Computers: Devices through end user interact with cloud.
2. Distributed Servers: Servers working together that are placed in a
geographical area.
3. Datacenters: Collection of servers where application is placed and accessed
via internet.

3
 INTRODUCTION(2/3)

• Cloud computing provides saleable on-demand services:

• Infrastructure as a Service(IaaS)
• IaaS is the delivery of technology infrastructure as an on demand scalable
service.
• Platform as a Service(PaaS)
• PaaS provides the runtime environment for applications, development &
deployment tools, etc.
• Software as a Service(SaaS)
• SaaS model allows to use software applications as a service to end users

4
 INTRODUCTION(3/3)

• A cloud virtualized resource is a set of specification and configuration files

called a Virtual Machine(VM)
• Due to dynamic nature workload of VM’s fluctuates dynamically.
• VM Migration is used to relax the workload by moving a VM from overloaded
Physical Machine(PM) to under loaded PM.
• VM Migration by suspend/resume strategy or live migration
• Proposed Multi-Objective Load Balancing for VM Migration

5
 RELATED WORK(1/2)

• Several methods have been proposed to migrate a VM from one physical

resource to another.
• ZAP System uses stop and copy strategy.[1]
• Jun and Xiaowei [2] developed a VM live migration policy for the IPv6 network.
• Pre-copy migration method is applied by vMotion component of VMware
vSphere

6
 RELATED WORK(2/2)

• Nicolae and Cappello[3] developed a strategy to store live snapshots of the

VM instance disk.
• Atif and Strazdins[4] developed a cloud visualization framework for Application
as a Service(AaaS)

7
 MULTI-OBJECTIVE LOAD
BALANCING(MO-LB) SYSTEM(1/2)

• A hotspot occurs when the performance of VM

degrades because the PM is unable to respond to
the user demand.
• When the resources of PM are under utilized it is
termed as cold spot.
• Both situations can be handled by moving one or
more VM’s from an overloaded VM to other VM.

8
 MULTI-OBJECTIVE LOAD BALANCING
SYSTEM(2/2)

• Steps in VM Load Balancing

1. Create Clusters of PM’s
With set of VM’s
2. Create set of VM’s to
deliver SaaS
3. Extra work is allocated to
compatible VM’s that are
allocated on other PM’s

9
 MO-LB SYSTEM MODEL(1/5)

• MO-LB System has four parts:

• Global blackboard (GB)
• CUP sub-system
• MOTS-PSO sub-system
• Decision Maker Center(DMC)

10
 MO-LB SYSTEM MODEL(2/5)

• Global blackboard (GB):

In GB all VMM’s and task schedulers share their data and information
about VM features , jobs and scheduled tasks in a cluster.
• CUP sub-system:
CUP sub-system predicts two situations, based on that DMC
determines weather the MO-LB system or VM migration should be applied.

11
 MO-LB SYSTEM MODEL (3/5)

• MOTS-PSO sub-system:
MOTS-PSO applies information applies all the information supplied by
other parts of framework. It considers:
1. Task transfer time
2. Task execution time
3. Execution cost
4. Power consumption

12
 MO-LB SYSTEM MODEL(4/5)

• DMC:
DMC transfers task from MOTS-PSO to the predetermined set of VM’s
based on suggested scheduling pattern.

13
MO-LB SYSTEM MODEL(5/5)

14
 PARTICLE SWARM OPTIMIZATION

• PSO is initialized with a group of random particles (solutions) and then

searches for optimal by updating generations.
• Particles move through the solution space, and are evaluated according to
some fitness criterion after each time step. In every iteration, each particle is
updated by following two "best" values.
• The first one is the best solution (fitness) it has achieved so far (the fitness
value is also stored). This value is called pbest.

15
 EVALUATION RESULTS(1/2)

CUP Results
VM id f CA¢cpuk( )x VMcpuk ( )ct VMloadk Number of (VMack )

1 ≥0 and ≥80 then Overloaded 0

5 ≥0 and ≥80 then Overloaded 0
3 ≤0 and ≤80 then Under-loaded 2
6 ≤0 and ≤80 then Under-loaded . 2
7 ≤0 and ≤80 then Under-loaded 2
9 ≤0 and ≤80 then Under-loaded 1
10 ≤0 and ≤80 then Under-loaded 1
11 ≤0 and ≤80 then Under-loaded 1

Optimal suggested pattern for scheduling Tse1 t to VM1set

Tasks t47 t49 t30 t21 t25 t27 t50
.
Task CPU usage (GHz) 67.166 67.166 21.186 21.186 21.186 21.186 67.166
VMs VM11 VM3 VM9 VM3 VM3 VM9 VM3

16
EVALUATION RESULTS (2/2)

Comparison of results.

17
 CONCLUSION AND FUTURE WORKS

• Several VM migration techniques have been applied for load balancing and
optimizing resource utilization in cloud environments,
• Live migration for large VMs is not an optimal solution because it consumes
time, bandwidth, power and memory space.
• MO-LB system has been proposed instead of VM migration for load balancing.
The MO-LB system distributes accumulated tasks in the task queue of the
primary VMs over a set of compatible VMs with lower utilization.
• Future study will also consider heterogeneous cloud environments as its
research direction, and the prediction sub-system of the developed model will
be improved.

18
 REFERENCES(1/2)

[1] Osman, S., Subhraveti, D., Su, G. and Nieh, J. (2002) The
design and implementation of ZAP: A system for migrating
computing environments. ACM SIGOPS Operat. Sys. Rev, 36,
361–376.
[2] Jun, C. and Xiaowei, C. (2011) IPv6 virtual machine live
migration framework for cloud computing. Energ. Proceed.,
13, 5753–5757.

19
 REFERENCES(2/2)

[3]Nicolae, B. and Cappello, F. (2013) BlobCR: Virtual disk

based checkpoint-restart for HPC applications on IaaS clouds.
J. Parallel Distrib. Comp., 73, 698–711.
[4]Atif, M. and Strazdins, P. (2014) Adaptive parallel application
resource remapping through the live migration of virtual
machines. Fut. Gen. Comp. Sys., 37, 148–161.

20