Message boards : The Lounge : Why not in TOP500
Message board moderation
Author | Message |
---|---|
Send message Joined: 10 Mar 08 Posts: 1 |
Why not in top 500? BOINC is a massive computer itself. And one of the world's biggest BOINC MUST be in the list, and in the TOP TEN, if i don't miss TERAFLOPS............ What do you believe? |
Send message Joined: 12 Feb 06 Posts: 232 |
TOP500 is measured with the LINPACK benchmarks. Nobody has tried those on BOINC. -- Eric Myers "Education is not the filling of a pail, but the lighting of a fire." -- William Butler Yeats |
Send message Joined: 3 Apr 06 Posts: 547 |
TOP500 is measured with the LINPACK benchmarks. Maybe it would be worth at least the next April Fools' day? :-) If some project could add e.g. Linpack 100x100 Benchmark In C/C++ For PCs as their application, correlate the results with Boinc's flops ant then try to extrapolate the comparison to the BOINC supercomputer teraflops? Maybe if Boinc alpha would use it instead of / in addition to Uppercase, we could get more alpha testers? ;-) (I know, actually a sort of Linpack Benchmark for Distributed-Memory Computers should be used for Boinc (not as easy to develop, test and distribute), which Linpack results would be for sure lower, but possibly more acceptable?) Peter |
Send message Joined: 12 Feb 06 Posts: 232 |
Peter wrote:
Or earlier. I started work on such an app for Pirates@Home, but it's been low on the priority list. But as you suggest, if we could run both a Linpack benchmark and the whetstone/drystone benchmarks which BOINC uses, and establish a significant statistical correlation between the two, that would give us a better idea of where BOINC projects might fit in the TOP500, even if they never get officially recognized. In reading the Linpack benchmark code I was surprised to learn that they solve the same system over and over again, instead of a variety of 1000x1000 matrices. So it would also be interesting to see how much variation there is between the specific case used for the TOP500 and a more realistic scenario.
The HPC benchmark is more appropriate to a Beowulf cluster where the nodes can share either memory or disk space between them. In contrast the nodes of a BOINC cluster are separated individuals, so the TPP benchmark is more appropriate. It might also be necessary to compare TPP and HPC on the same cluster to establish the correlation between the two. (For details see the "Linpack" article in Wikipedia, including the discsussion page.) -- Eric Myers "Education is not the filling of a pail, but the lighting of a fire." -- William Butler Yeats |
Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License,
Version 1.2 or any later version published by the Free Software Foundation.