A geometric model for multiple processor performance analysis

A theoretical model is proposed to investigate the effect of a job's parallel software structure on multiple-processor performance. Jobs are represented in terms of their software structure using a geometric model known as a job profile. Using three performance indices-efficiency, execution time, and speedup-a methodology is developed to evaluate the performance of processors executing a job modeled as a job profile. Equations for efficiency, execution time, and speedup are derived for an M-processor machine executing a job represented by a generalized job profile. These equations are analyzed for a specific job profile to determine what bounds are placed on performance by a job's software structure. The results indicate that the processor-sharing and processor-sweeping scheduling disciplines provide upper and lower bounds on the performance of a multiple-processor machine using a work-conserving scheduling policy.<>