Program analysis and scheduling for distributed computing on handheld devices
Abstract
Distributed systems for high performance computing (HPC) have been extensively studied in many years. However, little work has been done on distributed computing on handheld devices. Resource constraints on handheld devices and wireless networks make it difficult to use traditional distributed computing techniques in an environment where handheld devices are connected to stationary computers through wireless networks. In this thesis, we present a framework which combines static program analysis and run-time task scheduling to achieve efficient client-server distributed computing on handheld devices. In our framework, a client is a handheld device which is connected to a server through the wireless network. We analyze an ordinary program, divide its computation into tasks and then define an optimization problem over a task control flow graph (TCFG) for optimal task scheduling. Our cost analysis estimates the execution time and expresses it as a function of several run-time parameters. Our task scheduling algorithm then finds the optimal task scheduling corresponding to different ranges of run-time parameter values. Finally, the given program is transformed into a client-server distributed program based on our static analysis results. At run time, the distributed program self-schedules its task execution corresponding to the current run-time parameter values and guarantees correct program semantics during the distributed execution. We conduct experiments to measure both the program running time and the energy consumption using a set of benchmark programs on two handheld devices, one with a StrongArm processor and the other with an XScale processor. Experimental results show that our approach can significantly improve the performance and the energy consumption on handheld devices. Different run-time parameter values can lead to quite different task scheduling decisions.
Degree
Ph.D.
Advisors
Li, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.