Questions and Answers :
Macintosh :
SETI 5.22 very much slower than AlexKan 5.13
Message board moderation
Author | Message |
---|---|
Keith Send message Joined: 19 May 99 Posts: 483 Credit: 938,268 RAC: 0 |
I removed Alex Kan's optimized 5.13 and the xml file and reset. Great. All went well. There were no longer those -5000 messages. There were no longer work units (in fact none at all) pending with 2 successful tasks Then, I noticed the tasks were apparently very much larger, and were expected to take days to finish instead of about 15 hours maximum previously. (So, maybe the units were just simply being sent in larger chunks.) Strangely, it was later noted that: !. Tasks completed in about 10% less time than was originally shown. 2. Tasks were regularly granted 10% less credit than claimed. 3. A typical task was clocking up 30k secs cpu time for 27 credits and previously clocking up 9.6k secs and granted 59 credits. 22nd July was the cut-off point should you wish to look at the "Results for Computer" at the following link: http://setiathome.berkeley.edu/results.php?hostid=2881519 As this seemed to be consistently happening, I decided to revert to Alex Kan 5.13. Then, lo and behold, the current cache was resent (as was expected in BOINC 5.10.10 or 5.10.19) and the very same tasks that had been scheduled to take over 40 hours to crunch under v5.22 were now scheduled to complete in just 10 and 11.5 hours!! Keith |
Dotsch Send message Joined: 9 Jun 99 Posts: 2422 Credit: 919,393 RAC: 0 |
The 5.22 application is the new release which could use the multibeam data. So far I know is the data we analyse still multibeam data. But I am not shure, if the multibeam data takes longer to analyse. Alex application is a optimized version with changes in the code and some optimisations for the altivec of the PPC CPUs. Eric has announced a new version for tomorrow, which fix the most issues in 5.22. The 5.2x applications claims more credits as the old 5.1x applications. This is made internal in the application. If you WUs would be validated with older 5.1x results, you get the lesser credits. |
Keith Send message Joined: 19 May 99 Posts: 483 Credit: 938,268 RAC: 0 |
This does not answer my observation that Alex 2.13 seems to be taking a quarter of the cpu time to achieve twice the credit granted when compared to SETI 5.22. Something must be wrong here. Note the last part of my last post: 22nd July was the cut-off point should you wish to look at the "Results for Computer" at the following link: http://setiathome.berkeley.edu/results.php?hostid=2881519 As this seemed to be consistently happening, I decided to revert to Alex Kan 5.13. Then, lo and behold, the current cache was resent (as was expected in BOINC 5.10.10 or 5.10.19) and the very same tasks that had been scheduled to take over 40 hours to crunch under v5.22 were now scheduled to complete in just 10 and 11.5 hours!! Keith |
Dotsch Send message Joined: 9 Jun 99 Posts: 2422 Credit: 919,393 RAC: 0 |
The point is, that I am not know in the moment, if the 5.22 (MB) application does the full computation of the Multibeam data, so it is OK that the crunching times are longer, because a lot of more things must be computed compared with the old non MB application (5.13). Also, it is Ok that 5.22 clames more credits, see my last explanations. You can not compare 5.22 with 5.13 against the credits and the crunching times against each other, because 5.13 from Alex is strongly optimized at the CPU types and we not know in the moment how 5.22 utilize the MB data. Also a unkown factor is, how 5.22 using the differnet CPU capatilies (altivec for example). |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.