Online Performance Projection for Clusters with Heterogeneous GPUs

TitleOnline Performance Projection for Clusters with Heterogeneous GPUs
Publication TypeConference Paper
Year of Publication2013
Conference Name19th IEEE International Conference on Parallel and Distributed Systems
Conference LocationSeoul, KOR
Other NumbersANL/MCS-P5046-1213
Abstract

We present a fully automated approach to project the relative performance of an OpenCL program over different GPUs. Performance projections can be made within a small amount of time, and the projection overhead stays relatively constant with the input data size. As a result, the technique can help runtime tools make dynamic decisions about which GPU would run faster for a given kernel. Usage cases of this technique include scheduling or migrating GPU workloads over a heterogeneous cluster with different types of GPUs.

 

PDFhttp://www.mcs.anl.gov/papers/P5046-1213.pdf