| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: FightAIDS@Home Phase 2
Thread: Hughe runtime discrepancy between Linux and Windows |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 58
|
|
| Author |
|
|
BobCat13
Senior Cruncher Joined: Oct 29, 2005 Post Count: 295 Status: Offline Project Badges: 
|
Small sample size on a dual boot Core2 Q6600 @ 3.20 GHz
LinuxMint 17.2 (2 tasks) 23.64 hrs 23.10 23.37 avg Win Vista (4 tasks) 16.00 hrs 16.03 16.07 16.09 16.05 avg |
||
|
|
3rkko
Advanced Cruncher Finland Joined: Aug 2, 2008 Post Count: 105 Status: Offline Project Badges:
|
The nerds who develope the programs usually use Linux, and only later convert the programs to Windows/Mac, which makes Linux the fastest platform most of the time. We the Windows majority just seem to be lucky this time around.
----------------------------------------PS. i7 & i3 @ 3.5ghz takes about 12 hours / WU on Win10.  [Edit 1 times, last edit by 3rkko at Oct 3, 2015 7:59:25 PM] |
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges:
|
The nerds who develope the programs usually use Linux, and only later convert the programs to Windows/Mac, which makes Linux the fastest platform most of the time. We the Windows majority just seem to be lucky this time around. PS. i7 & i3 @ 3.5ghz takes about 12 hours / WU on Win10. maybe not so lucky? maybe this time they developed it on Win platform...after all there is a majority of people on Win!  |
||
|
|
branjo
Master Cruncher Slovakia Joined: Jun 29, 2012 Post Count: 1892 Status: Offline Project Badges: 
|
The same issue is with OSX. Seems this app is not optimized for Unix clones (as opposite to VINA ones) but for me it is OK as far as the results are valid.
----------------------------------------Cheers   Crunching@Home since January 13 2000. Shrubbing@Home since January 5 2006  |
||
|
|
Rickjb
Veteran Cruncher Australia Joined: Sep 17, 2006 Post Count: 666 Status: Offline Project Badges: 
|
I too noticed the slow FAHB throughput under Linux, and have switched all of my Linux crunching to VINA (OET + occasional FAHV resends). This should actually increase the throughput of FAHB for WCG as a whole, vs me continuing to crunch FAHB under either Linux or Windows. By crunching OET under Linux, I'm taking a boosted share of OET away from being crunched by the Windows majority, thus leaving more of other members' Windows machines to crunch FAHB.
Well ... that's how I justify going for all the extra credit I get for running OET under Linux, anyway. --- There are a number of possible explanations for the difference in FAHB performance between Windows and Linux, e.g.
Please add my voice to the calls for WCG to investigate, explain & hopefully fix the discrepancy in FAHB performance between Unix clones & Windows. |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Did anyone speak up during Beta? IDNR. There's longer history of a first pass launch having dramatic platform performance differences. Best recollected is HCC. It started off something like 2-4x as slow on *nix and whence optimized, ended up twice as fast, just like AD Vina, in the HCC case removal of redundant loops or something. The choice was facing wine [sic] of FAHB not being available at launch for *nix, [or do I have this wrong ;?]. Yes it could be postponed with statement, but then would there have been performance statistics on a larger data set?
Anyway, on the dry part, think they've not yet switched on the 2nd gen work creator... WCG will have wanted serious data before doing so. (One day we will get "What we will do", but that is maybe too in the [new to me] SCI department. ;P On the loops, there was one during testing which got activated during restart... maybe a regression hardcode on the *nix build [all guess and speculation channels opened  ] |
||
|
|
Eric_Kaiser
Veteran Cruncher Germany (Hessen) Joined: May 7, 2013 Post Count: 1047 Status: Offline Project Badges: 
|
Rob, I've seen the discrepancy in the Beta as well and yes, I didn't mentioned it there. Perhaps I thought that the release into production wasn't so close as it was in the end.
----------------------------------------But it's never to late to address it. And wine isn't an option on headless servers...  |
||
|
|
Jason1478963
Senior Cruncher United States Joined: Sep 18, 2005 Post Count: 295 Status: Offline Project Badges: 
|
For me the runtimes are one issue, but the points awarded for windows vs linux is far more disturbing.
----------------------------------------Examples: Windows machine: Runtime: 12.05 / 12.20 Points 390.2 / 390.2 10.95 / 11.07 Points 353.6 / 353.6 Vs. Linux: 39.23 / 39.29 Points 384.2 / 384.2 39.50 / 39.56 Points 384.2 / 384.2   |
||
|
|
BobCat13
Senior Cruncher Joined: Oct 29, 2005 Post Count: 295 Status: Offline Project Badges:
|
For me the runtimes are one issue, but the points awarded for windows vs linux is far more disturbing. The first 2 beta tasks run under linux were awarded approximately 465 credits, but since then all linux tasks (beta and production) have gotten exactly 384.2 credits no matter the runtime. Even more interesting is this from a Windows machine: FAH2_ avx101122_ 000076_ 0001_ 002_ 0-- Green-Scream Valid 10/3/15 00:30:47 10/3/15 22:02:52 21.13 / 21.39 502.4 / 502.4 FAH2_ avx17684-ls_ 000008_ 0010_ 001_ 0-- Green-Scream Valid 10/3/15 05:53:09 10/4/15 01:29:05 19.30 / 19.58 329.4 / 329.4 Two tasks reported less than 3.5 hours apart are rewarded very differing credit amounts per hour, 23.77 c/h vs. 17.06 c/h. |
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges:
|
I too noticed the slow FAHB throughput under Linux, and have switched all of my Linux crunching to VINA (OET + occasional FAHV resends). This should actually increase the throughput of FAHB for WCG as a whole, vs me continuing to crunch FAHB under either Linux or Windows. By crunching OET under Linux, I'm taking a boosted share of OET away from being crunched by the Windows majority, thus leaving more of other members' Windows machines to crunch FAHB. Well ... that's how I justify going for all the extra credit I get for running OET under Linux, anyway. & that seems like a BEST solution 4 now! 2 be energy efficient & crunch Linux on those projects that r faster on those machines! |
||
|
|
|