TimeGraph: GPU scheduling for real-time multi-tasking environments
USENIXATC'11 Proceedings of the 2011 USENIX conference on USENIX annual technical conference
Hi-index | 0.00 |
This paper will report our evaluation to use openCL as a platform for hard realtime scheduling. Specifically, we have evaluated which types of tasks are faster on GPGPU than on CPU. We have investigated computational tasks, memory intensive tasks (especially tasks using low latency GDDR memory) and disk intensive tasks. This study is the first part of a larger research program to design an innovative Linux scheduler subsystem that runs on GPGPU and schedules tasks running on GPGPU as well as on CPU. Based on the results obtained from benchmarking various types of tasks, we found out that some of them are faster on GPGPU than on CPU and therefore should preferably be executed on GPGPU. Preliminary data suggest that we can expect a speed up of up to 10-fold with respect to execution time and latency.