چكيده به لاتين
Abstract
Nowadays, virtualization technologies are experiencing a renewed interest for improving energy effeciency and resource utilization as well as reducing costs. Besides, they are key technology for cloud computing and server consolidation. In spite of these benefits, virtualization technologies have posed some challenges including performance challenges. First of all, application performance running in virtualized environments can be degrade due to the overhead of hypervisor and the functionality of other virtual machines. Additionally, they increase system complexity which make it more difficult to predict and model the performance of applications in virtualized environments.
To deal with the mentioned challenges, in this thesis, we measured the performance of Web server in Xen environment under different scenarios and model it using Queueing Network Theory. Firstly, we measured the utilization of CPUs under different scenarios in Xen and showed that for a specific utilization boundary, as the request rate increases, the utilization of CPUs increases linearly. Therefore, the service demand of requests remains constant for different rates and CPUs act as a load-independent resources.
Secondly, we illustrated that in a single VM environment, the utilization of CPU pinned to DomU is equal for both PV and HVM modes whereas the utilization of CPU pinned to Dom0 that handles the I/O of DomU is higher in HVM mode compared with PV mode. This indicates that HVM mode has higher I/O overhead on Dom0.
Thirdly, we investigated the effect of packet size on the service demand of requests at CPUs in multi VM environmentand showed that for small packet sizes, the service demand at CPUs pinned to DomUs is higher in HVM mode compared with PV mode while for big packet sizes, the service demands of both modes are equal. In addition, for all packet sizes employed in the experiments, the service demand of HVM mode on CPU pinned to Dom0 is higher than PV mode.
Furthermore, for each scenario, we provided a queueing network model, calculated system response time using model, and compared the results from model with measured ones in order to verify proposed model.
Keywords: Virtual Machines,Xen Hypervisor, Performance Measurement, Performance Modeling, Queueing Network Model
