Resource management for cloud data centers

Abstract

With the rapid growing number of Cloud applications, demands for large-scale data centers have raised to historical high. Technology advancements in recent years make it possible to manufacture high performance processors and server units at low cost. While it is feasible to have thousands of processors in a data center, the associated energy problems can be catastrophic. High-energy consumption contributes to high operational cost, unbalanced temperature distribution, and high hardware failure rates at data centers. Therefore, there is an urgent need for developing efficient operational schemes for Cloud data centers. Cloud data centers allow dynamic and fexible resource provisioning to accommodate time varying computational demands. To maximize resource utilization, Cloud service providers employ dynamic virtual machine (VM) migrations technologies. This thesis aims to present different VM consolidation mechanisms for better resource management in Cloud data centers. First, a thermal-aware VM consolidation mechanism is proposed for resource allocation optimization and server reliability assurance in Cloud data centers. The proposed mechanism takes both host power consumption and temperature into account. The variability in host temperature, which has been shown to have a negative impact on server reliability, is considered as a migration criterion during the consolidation process. A Markov model is further adopted to predict future CPU usages of physical hosts and VMs to reduce the number of migrations needed in the long run. Performance parameters, including energy consumption, Service Level Agreements (SLA) violations with outage, number of outage incidents, and migration number, are evaluated. Then, a VM consolidation mechanism inspired by host-switching behaviors in symbiotic associates is proposed. Two heuristic functions which have been inspired by host susceptibility and symbiotic coefficient among symbionts, are proposed to deal with utilization levels of hosts and resource utilization correlations among co-located VMs. In order to hedge the risk of host overloading, VMs having low symbiotic coefficient values will not be assigned to a host which is regarded as susceptible in the symbiosis analogy. The performance of the proposed bio-inspired heuristics based mechanism is compared with other existing correlation-based VM allocation mechanisms. Moreover, experiment results are analyzed and discussed. Finally, the VM allocation problem is further formulated as a stable matching problem. A deferred acceptance procedure is adopted to resolve conficts among VMs and physical hosts. During the matching process, each VM ranks the hosts according to their maximum correlation level after migration to preserve the quality of service. Similarly, each host has its own preference list regarding a combination of VMs such that the host can operate close to a desirable utilization threshold. The proposed VM consolidation mechanism can effectively reduce energy consumption and minimize violations of SLA in Cloud data centers.