Uniform TitleEvaluating the performance characteristics of a virtual machine used on simultaneous multi-threaded (SMT) processors
NameYim, Hiu Shan (author), Pompili, Dario (chair), Silver, Deborah (internal member), Gruteser, Marco (internal member), Rutgers University, Graduate School - New Brunswick,
SubjectElectrical and Computer Engineering,
Parallel processing (Electronic computers)
DescriptionVirtualization of computing hardware is one technique which can make possible the use of fewer physical computers, thus lowering resource consumption. Today, as in the past, hardware performance remains a major bottleneck to virtual machine performance. Simultaneous multithreaded, or SMT, processors provide thread-level parallelism and are being used to overcome the performance limitations of virtual machines. These same processors are also being used to decrease the cost of computing systems since less hardware and power is required when compared with multiple CPU systems. Virtual machines should benefit from the properties of SMT processors since they have a common cache and parallel execution threads. As a result, using virtual machines in combination with SMT processors should be an efficient way to maintain or increase performance, save money and reduce physical hardware requirements. This study attempts to determine if an improvement on virtual machine performance exists through the use of an SMT processor. If the performance of an SMT processor-based system is on-par with several independent computers or multiple CPU systems, then the use SMT would be an efficient way for organizations to achieve their performance requirements at a reduced cost. This thesis evaluates the performance of a virtual machine used with and without SMT. This study shows that a definite, measurable performance improvement exists when a virtual machine is run with an SMT processor and that better virtual machine performance is achieved as load increases. Then a performance modeling method is suggested for various combinations of SMT and virtual machines in order to predict and maximize system performance and achieve proper load balancing.
NoteIncludes bibliographical references (p. 51-53).
CollectionGraduate School - New Brunswick Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.