Optimal resource allocation and migration decision for virtual machine requests in cloud data centers

Date

2023-8-29

Author

Mert, Nazım Kerem

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

123
views

6
downloads

Cite This

Infrastructure as a Service (IaaS) in cloud computing utilizes virtualization technology to instantiate Virtual Machines (VMs) on the Physical Machines (PMs) of a Cloud Data Center (CDC) based on user requests. In this context, the Service Level Agreement (SLA) with the CDC user defines the allocated resources for a VM in CPU cores, Memory, Disk capacity, and network bandwidth. The resource allocation should provide the demanded resources without an SLA violation (SLAV), use the PM resources efficiently and minimize the power consumption. The PMs can have different power ratings, running VMs with different utilizations. To this end, deciding the PM to instantiate a VM affects both the performance and the cost in terms of power and allocating resources. In this thesis, we propose, formulate, implement, and evaluate a new resource allocation method for CDCs that we call CLOUDMAN++. CLOUDMAN++ accounts for the instantaneous utilization of the running VMs and allocates the unused resources to new VM requests. On the one hand, this approach is more resource and power efficient as it turns on less number of PMs. On the other hand, if the VMs increase their utilization, the PM might reach a total of $100\%$ utilization resulting in SLAVs. We resolve this by migrating VMs on such PMs. CLOUDMAN++ is an ILP formulation with a cost function to minimize power consumption and migrations. The ILP formulation decides on the PMs to place the new VM requests, the VMs to be migrated if necessary, and their target PMs. As a special feature, the formulation offers a number of tunable design parameters. We evaluate CLOUDMAN++ using an actual CDC workload trace which provides VM requests and their instantaneous/dynamic utilization. We first perform a sensitivity analysis to find the values of the design parameters to use with the trace at hand. We then compare CLOUDMAN++ with another ILP-based resource allocation algorithm and a legacy algorithm that we extend with migration features. The results show that CLOUDMAN++ achieves more efficient resource use by exploiting the underutilized but allocated resources while resulting in better service to the users. Furthermore selecting the values of the design parameters according to the workload characteristics at hand significantly affects the performance of the resource allocation.

Subject Keywords

Cloud data center, Resource management, Virtual machine, Migration, Dynamic resource allocation

URI

https://hdl.handle.net/11511/105388

Collections

Graduate School of Natural and Applied Sciences, Thesis