الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 93 |  |  |
| | | from | 93 | | |
Abstract
The wide adoption of cloud computing is leading to an increase in the number of data centers and physical servers worldwide. Eco-logical communities are calling for the data centers to ”go green” and hence, energy e ciency has become a crucial concern in modern data centers. Dynamic virtual machine (VM) consolidation is one of the e ective approaches endorsed to achieve energy e ciency in cloud data centers hosting thousands of servers. Live migration is a core feature enabling virtual machine consolidation. However, live migration is a costly operation imposing energy and performance overhead. Thereby, an e cient dynamic virtual machine consolidate should consider the cost due to live migration.
This thesis addresses this challenge by presenting a dynamic vir-tual machine consolidation algorithm that is live migration-overhead aware. The dynamic VM consolidation problem is approached as a discrete combinatorial bi-objective problem of saving the most en-ergy while keeping the migration cost at minimum. Factors a ecting live migration cost and the parameters contributing to that cost are studied and an estimation model for migration overhead is proposed. Thereafter, a dynamic VM consolidation algorithm is designed and implemented based on a meta-heuristic algorithm, simulated an-nealing, that accounts for the migration cost imposed by a given consolidation plan. Simulation-based experiments are conducted on
iii
CloudSim using real cloud workload traces from PlanetLab to eval-uate the performance of the proposed algorithm. Results of the proposed algorithm are compared to those of Best Fit, First Fit and Worst Fit heuristic algorithms. For each experiment, the amount of energy consumed, number of consolidated servers and the number of carried out migrations are tracked. Among the compared algorithms, it was found that the migration-unaware rst t approach provides the least data center power consumption as well as number of re-leased physical machines. The proposed migration overhead-aware simulated annealing based algorithm is found to consume almost the same amount of energy of that using a First Fit based consolida-tion. However, the proposed algorithm accounts for the cost due to live migration and is shown to reduce number of performed VM migrations by 10% compared to FF-based algorithm. . . .