Resource management in heterogeneous computing systems: Continuously running applications, tasks with priorities and deadlines, and power constrained mobile devices
Abstract
In a distributed heterogeneous computing (HC ) system, the resources (computing machines and communication links) have different capabilities (e.g., different machines may have different processing speeds) and tasks (e.g., an execution of an application) have different needs (e.g., a program may need to process integer numbers). To maximize the performance of the system, it is essential to map tasks onto machines (to assign resources to tasks and order the execution of tasks on each resource) in a manner that exploits the heterogeneity of the resources and tasks. The mapping of tasks onto machines in an HC environment has been shown, in general, to be an NP-complete problem. Static mapping heuristics are used in an off-line planning phase. Dynamic mapping is performed when the arrival of tasks are not known a priori. Three environments are investigated in this research. The first is an environment with continuously running communicating applications and heterogeneous machines. The goal is to find an initial static allocation of applications onto heterogeneous machines to maximize the allowable workload increase until a dynamic reallocation of resources is required. The second environment studies a dynamic mapping problem with tasks having priorities and multiple deadlines. The objective is to complete as many tasks as possible to achieve the most value accrued during an interval of time (value of a task depends on the priority of the task and when the task is completed). The third environment is a dynamic ad hoc grid environment where: (a) the tasks have weighted priorities and deadlines, the tasks need inputs from other sources, and results must be sent to the task request originator machine; (b) the clock rate of each machine can vary depending on the rate of energy consumption for that machine at any point in time (dynamic voltage scaling); (c) the battery capacity of each machine is limited; and (d) the system is expected to operate for eight hours. The goal of this research was to maximize a performance measure based on the weighted priorities of the tasks that completed by their deadlines within the eight-hour session.
Degree
Ph.D.
Advisors
Eigenmann, Purdue University.
Subject Area
Electrical engineering|Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.