Name
Abstract | ||
|---|---|---|
Resource management on accelerator based systems is com- plicated by the disjoint nature of the main CPU and acceler- ator, which involves separate memory hierarhcies, di! erent degrees of parallelism, and relatively high cost of communi- cating between them. For applications with irregular par- allelism, where work is dynamically created based on other computations, the accelerators may both consume and pro- duce work. To maintain load balance, the accelerators hand work back to the CPU to be scheduled. In this paper we consider multiple approaches for such scheduling problems and use the Cell BE system to demonstrate the di! erent schedulers and the trade-o! s between them. Our evaluation is done with both microbenchmarks and two bioinformatics applications (PBPI and RAxML). Our baseline approach uses a standard Linux scheduler on the CPU, possibly with more than one process per CPU. We then consider the ad- dition of cooperative scheduling to the Linux kernel and a user-level work-stealing approach. The two cooperative ap- proaches are able to decrease SPE idle time, by 30% and 70%, respectively, relative to the baseline scheduler. In both cases we believe the changes required to application level codes, e.g., a program written with MPI processes that use accelerator based compute nodes, is reasonable, although the kernel level approach provides more generality and ease of implementation, but often less performance than work stealing approach. |
Year | DOI | Venue |
|---|---|---|
2009 | 10.1145/1531743.1531769 | Conference On Computing Frontiers |
Keywords | Field | DocType |
dynamic parallelism,cell be,main cpu,multiple approach,use accelerator,accelerators hand work,kernel level approach,cooperative approach,baseline approach,cooperative scheduling,user-level work-stealing approach,application level code,linux kernel,scheduling problem,resource manager,load balance | Resource management,Kernel (linear algebra),Central processing unit,Fair-share scheduling,Computer science,Scheduling (computing),Load balancing (computing),Parallel computing,Real-time computing,Work stealing,Operating system,Linux kernel | Conference |
Citations | PageRank | References |
1 | 0.42 | 18 |
Authors | ||
6 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Filip Blagojevic | 1 | 431 | 24.38 |
| Costin Iancu | 2 | 529 | 34.78 |
| Katherine A. Yelick | 3 | 3494 | 407.23 |
| Matthew Curtis-Maury | 4 | 324 | 19.19 |
| Dimitrios S. Nikolopoulos | 5 | 1469 | 128.40 |
| Benjamin Rose | 6 | 66 | 4.08 |