Find in Library

Currently, the <i>User-PC computingsystem (UPC)</i> has been studied as a low-cost and high-performance distributed computing platform. It uses idling resources of personal computers (PCs) in a group. The job-worker assignment for minimizing <i>makespan</i> is critical to determine the performance of the UPC system. Some applications need to execute a lot of <i>uniform jobs</i> that use the identical program but with slightly different data, where they take the similar CPU time on a PC. Then, the total CPU time of a worker is almost linear to the number of assigned jobs. In this paper, we propose a <i>static assignment algorithm</i> of <i>uniform jobs</i> to workers in the UPC system, using <i>simultaneous linear equations</i> to find the <i>lower bound</i> on <i>makespan</i>, where every worker requires the same CPU time to complete the assigned jobs. For the evaluations of the proposal, we consider the uniform jobs in three applications. In <i>OpenPose</i>, the CNN-based <i>keypoint</i> estimation program runs with various images of human bodies. In <i>OpenFOAM</i>, the physics simulation program runs with various parameter sets. In <i>code testing</i>, two open-source programs run with various source codes from students for the <i>Android programming learning assistance system (APLAS)</i>. Using the proposal, we assigned the jobs to six workers in the testbed UPC system and measured the CPU time. The results show that <i>makespan</i> was reduced by <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>10</mn><mo>%</mo></mrow></semantics></math></inline-formula> on average, which confirms the effectiveness of the proposal.