Speedup and Scalability of SPMD Computations on Workstation Clusters.
Kemal Efe, Putchong Uthayopas, Adrienne L. Broadwater · 1999
Scalabiltity is among the most important advantages attributed to workstation clusters. A new application may be initially started with a few workstations, and new workstations may be added as demand increases for computing power. However, designing scalable systems requres finetuning of several hardware and software parameters. This paper introduces the maximum-speedup scalability as a performance metric to be used when evaluating design alternatives for a hardware /software combination, and demonstrates the model with measurements taken from real implementations. Keywords: Workstation clusters, scalability, speedup, parallel computing 1 Introduction Ideally, we would like to double the speed of parallel computations when we double the number of processors, but this is not possible with the exception of a few trivial cases. Often, parallel speed depends on a complex interplay between the system architecture and software, as well as special parameters arising from the hardware. Beyon...