The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	High performance cloud computing on multicore computers
Author:	Shan, Jianchen
View Online:	njit-etd2018-030 (xv, 132 pages ~ 1.2 MB pdf)
Department:	Department of Computer Science
Degree:	Doctor of Philosophy
Program:	Computer Science
Document Type:	Dissertation
Advisory Committee:	Ding, Xiaoning (Committee chair) Borcea, Cristian (Committee member) Gehani, Narain (Committee member) Curtmola, Reza (Committee member) Liu, Qing Gary (Committee member)
Date:	2018-05
Keywords:	Cloud computing Computer architecture High performance computing Multicore Operating system Virtualization
Availability:	Unrestricted
Abstract:	The cloud has become a major computing platform, with virtualization being a key to allow applications to run and share the resources in the cloud. A wide spectrum of applications need to process large amounts of data at high speeds in the cloud, e.g., analyzing customer data to find out purchase behavior, processing location data to determine geographical trends, or mining social media data to assess brand sentiment. To achieve high performance, these applications create and use multiple threads running on multicore processors. However, existing virtualization technology cannot support the efficient execution of such applications on virtual machines, making them suffer poor and unstable performance in the cloud. Targeting multi-threaded applications, the dissertation analyzes and diagnoses their performance issues on virtual machines, and designs practical solutions to improve their performance. The dissertation makes the following contributions. First, the dissertation conducts extensive experiments with standard multicore applications, in order to evaluate the performance overhead on virtualization systems and diagnose the causing factors. Second, focusing on one main source of the performance overhead, excessive spinning, the dissertation designs and evaluates a holistic solution to make effective utilization of the hardware virtualization support in processors to reduce excessive spinning with low cost. Third, focusing on application scalability, which is the most important performance feature for multi-threaded applications, the dissertation models application scalability in virtual machines and analyzes how application scalability changes with virtualization and resource sharing. Based on the modeling and analysis, the dissertation identifies key application features and system factors that have impacts on application scalability, and reveals possible approaches for improving scalability. Forth, the dissertation explores one approach to improving application scalability by making fully utilization of virtual resources of each virtual machine. The general idea is to match the workload distribution among the virtual CPUs in a virtual machine and the virtual CPU resource of the virtual machine manager.