Performance Evaluation of Cloud Services Considering Live Virtual Machine Migration

Nowadays, using virtualization techniques, users can access their private customized environment, while the service providers utilize their systems. Consequently, resource allocation based on users’ request and considering green computing is one of the most important research works. As a matter of fact, for every provider, it is economical to consolidate all the virtual machines and turn off unused physical servers. One of the main approaches which allows service providers to achieve this goal is Live Virtual Machine Migration. Live VM migration is used when you want to migrate a running VM from one physical server to another physical server, without disruption of VM execution. Although this approach is advantageous, it brings a couple of difficulties into the work. In this regard, we are trying to model this procedure with Petri net, more precisely, by using color Petri net we model an IaaS environment, represented by a large-scale data center consisting of N heterogeneous physical nodes. Web service providers submit their web-server application, and after getting required infrastructure, they can provide services to the client requests. In the figure below, each physical machine contains a virtual machine which is provisioned for a web-server application. After deploying and running the web-server application, the client can send their requests to the desired web-server. As the workload of each web-server is completely depends on the incoming request’s rate, we can investigate the effect of consolidation and live migration while encountering SLAV.

ehsan-etesami-thesis

The system provider wants to deliver service based on the SLA while minimizing power consumption. The novelty of this work is that we model the system using Petri net, in which each token can have a specific type (color set). Indeed the tokens of the system that represent physical hosts are distinguishable, and therefore dealing with live migration would be possible. In order to evaluate the proposed model, I calculated the power consumption, SLAV and drop probability of the proposed model and then verified results using CloudSim framework.

if you would like to know more please contact me via eetesami [-a@t-] ce.sharif.edu