Questions and Answers :
Unix/Linux :
Double Wammy
Message board moderation
Author | Message |
---|---|
Rosserver Send message Joined: 31 Aug 00 Posts: 11 Credit: 3,108 RAC: 0 |
Given; the Linux client seems to be about 35% slower than the windows client. And; the credits are awarded based on the time taken by the quickest of 3 participants to finish the WU. It would then be true that all WU's completed by some combination of linux and windows clients will then penalize the Linux user further by giving him less credit than the time actually worked. We work considerably longer for fewer credits. Could we just have some kind of communication on why the Linux client program is so slow, and if it can't be fixed then some kind of adjustment to the cobblestones? We all know that Linux is faster! |
UWP (Udo Wolter) Send message Joined: 28 Oct 02 Posts: 12 Credit: 12,105,371 RAC: 0 |
Yes, Linux ist faster but it depends on how good the program is written. It seems that the seti client itself is not really optimized, even compiling with optimization brings just 5% more speed. There is one thing you can do. As you said right, the credits are calculated by the speed but it's not the only thing. There's also the integer & floating point measuring which comes into this calculation too. In this case you can optimize the boinc client by just compiling it from source and using as much optimization as possible. I doubled my floating points and increased my integer calculations about 50%. In this case the work unit might last as long as before until it's finished but the calculation gets up straight. At least linux machines are hanging only 25% behind the windows machines if you use this method. But this 25% less speed is something what should be done in the seti client. As long as it is that poor optimized we will stay behind the windows machines...:( Mermgfurt, Udo |
Crunch3r Send message Joined: 15 Apr 99 Posts: 1546 Credit: 3,438,823 RAC: 0 |
That´s true I also recognised that after compiling boinc and the core client with the gcc flags CFLAGS="-O3 -march=athlon-tbird -fomit-frame-pointer -mmmx -m3dnow -fexpensive-optimizations" increased performance. But i also recogniced that credits depend on the memory installed. I´ve tested it on a RedHat Enterprise Linux running kernel 2.4.21. Two weeks with 256 mb ddram on athlon axp 2800 @ 2.2 Ghz average credit 30-40. Switching to 512 mb ddram i got 60 to 70 average cobblestones and rising. That´s courious... i don´t know why.... :) |
©2025 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.