Abstract | ||
---|---|---|
Accurate, reproducible and comparable measurement of collective operations is a complicated task. Although Different measurement schemes are implemented in well- known benchmarks,many of these schemes introduce differ- ent systematic errors in their measurements. We character- ize these errors and select a window-based approach as the most accurate method. However, this approach complicates measurements significantly and introduces a clock synchro- nization as a new source of systematic errors. We analyze approaches to avoid or correct those errors and develop a scalable synchronization scheme to conduct benchmarkson massively parallel systems. Our results are compared to the window-based scheme implemented in the SKaMPI bench- marks and show a reduction of the synchronization over- head by a factor of 16 on 128 processes. |
Year | DOI | Venue |
---|---|---|
2008 | 10.1109/IPDPS.2008.4536494 | Miami, FL |
Keywords | Field | DocType |
application program interfaces,message passing,parallel processing,synchronisation,SKaMPI benchmark,clock synchronization,massively parallel system measurement,performance analysis,window-based scheme,MPI,benchmarking,collective operations,scalable synchronization,time synchronization | Synchronization,Systematic error,Computer science,Parallel computing,Clock synchronization,Concurrent computing,Open system (systems theory),Message passing,Benchmarking,Scalability,Distributed computing | Conference |
ISSN | ISBN | Citations |
1530-2075 E-ISBN : 978-1-4244-1694-3 | 978-1-4244-1694-3 | 24 |
PageRank | References | Authors |
1.34 | 18 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Torsten Hoefler | 1 | 2197 | 163.64 |
Timo Schneider | 2 | 312 | 18.39 |
Andrew Lumsdaine | 3 | 2754 | 236.74 |