| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: BOINC credits vs WGC |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 109
|
|
| Author |
|
|
Johnny Cool
Ace Cruncher USA Joined: Jul 28, 2005 Post Count: 8621 Status: Offline Project Badges: 
|
Almost doubtlessly a .sh install versus a Synaptic install and then 32 bit weighing up against a 64 bit client. Found the 32 bit .sh benchmarks dramatically lower on integer. The Whetstone/floating point seems to only marginally vary, which is apparently dictated by the fact that those registers are 32 bit in hardware. When double, it's the Dhrystone/Integer. 6800 for Windows 64 bit, 12850 for 64 bit linux and 6500 if the package is used and if I remember well the 32 bit Vista gave about 4400, which then makes it nearly triple on Linux with the 64 bit Synaptic kit. Ludacris... but then HCC does go double speed which does not translate in double credit per hour. Anyway, I've now had a whole bunch where the low claiming wingman was dismissed and my quad claim was declared canonical... consistent very close to historic :P (keeping an eye out for the next scheduled benchmark and if reverting will rerun the script to half the iops rating. Someone on the other side will probably be okay with those bonuses ;o) (keeping an eye out for the next scheduled benchmark and if reverting will rerun the script to half the iops rating. What script and how?  Also, which version of Boinc are you using? Sorry if I overlooked what os you are running (64-bit?) I'm using the 64-bit of 6.10.58 for my 64-bit W7 Ultimate. I also downloaded ProcessLassoSetup. Was that a mistake to install?  Ehhh, I guess I will find out if it helps with the HCC project that I am running (as I am getting unusual low credit for my i7 860). Getting many results though!  Edit: just one example ... X0000034490937200407191810_ 1-- 608 Valid 8/14/10 11:22:16 8/14/10 15:42:45 2.63 33.7 / 33.7 X0000034490937200407191810_ 0-- 608 Valid 8/14/10 11:22:03 8/15/10 16:25:51 2.46 60.5 / 33.7 <----- moi ---------------------------------------- [Edit 1 times, last edit by Johnny Cool at Aug 15, 2010 5:48:27 PM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
We're talking Linux which has a (to me) excessive Dhrystone benchmark issue for certain compiles. Your client issue is more likely rooted in the HT sphere... takes your HT quad 2.46 hours, takes my non HT Q6600 Linux 64 1.20 hours on average.
----------------------------------------The Process Lasso ** on Windows works specifically for those that run ZR sciences... HFCC/FAAH and maybe C4CW too i.e. have a mini benchmark inside that impacts claim and grant. HCC is the none plus ultra integer crunching science, the high Dhrystone on Linux suggesting and measurement of run time confirming they run twice as fast on a Linux 64 client then the equiv.Windows i.e. 4x+ faster that the old version. ** It can do many things [bad things too if you don't know what you're doing], but that you'll have to discover for yourself.... you can have all 8 sciences run off 1 core with it if you want... that will really inflate run times and slash credit to probably always being the one of the wingman becoming the choice... an I7-860 getting 4 credit per hour  
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Johnny Cool
Ace Cruncher USA Joined: Jul 28, 2005 Post Count: 8621 Status: Offline Project Badges:
|
We're talking Linux which has a (to me) excessive Dhrystone benchmark issue for certain compiles. Your client issue is more likely rooted in the HT sphere... takes your HT quad 2.46 hours, takes my non HT Q6600 Linux 64 1.20 hours on average. The Process Lasso ** on Windows works specifically for those that run ZR sciences... HFCC/FAAH and maybe C4CW too i.e. have a mini benchmark inside that impacts claim and grant. HCC is the none plus ultra integer crunching science, the high Dhrystone on Linux suggesting and measurement of run time confirming they run twice as fast on a Linux 64 client then the equiv.Windows i.e. 4x+ faster that the old version. ** It can do many things [bad things too if you don't know what you're doing], but that you'll have to discover for yourself.... you can have all 8 sciences run off 1 core with it if you want... that will really inflate run times and slash credit to probably always being the one of the wingman becoming the choice... an I7-860 getting 4 credit per hour Sek, thanks for replying. I uninstalled the Process Lasso after checking it out for a few hours and I'm not sure it did anything positive as I am currently crunching on HCC. What the heck is C4CW? always being the one of the wingman becoming the choice... an I7-860 getting 4 credit per hour You horrible person you!  j/k |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Quad doing a Cameo as P3
----------------------------------------HFCC_ n1_ 00624739_ n1_ 0001_ 0-- 611 Valid 8/15/10 19:46:46 8/17/10 11:53:14 4.99 90.6 / 61.6 Too depressing these ZR sciences. 5 minutes 'assuming' the performance for the other 4.93 hours. Was compiling... yet those 4.93 hours of CPU cycles are still as were they having full exclusivity... that's what's being counted. :| Now I'm taking a rain-check before things start flying through the study, out the window on the ingombranti heap. 800900911, numero verde for environmentally dangerous objects.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
nasher
Veteran Cruncher USA Joined: Dec 2, 2005 Post Count: 1423 Status: Offline Project Badges: 
|
I have been messaging various members of my team to figure out why they used to crunch WCG and have lessened there crunching / switched off WCG.
----------------------------------------it seems that the major reason people have given up is the credits here are not as good as they are on other projects per cpu hour and the ones who have lessened are mostly people who are doing GPU crunching as well. in factboinc credit compair chart on the boincstats chart (click on the individual numbers to see the grapic representation there are only 5 projects out there that give LESS than WCG dose and over 55 that give more. if we increased the credits / WU by just a little bit (those given to BOINC) i think we would find an increase in crunching power here.. and from discussions on other projects that give on the high end or used to give oh the HIGH end (freeHAL). you can see that other complain if credits per hour are high but not if they are low.  |
||
|
|
anhhai
Veteran Cruncher Joined: Mar 22, 2005 Post Count: 839 Status: Offline Project Badges: 
|
I just converted one of my pcs over to Linux (dual boot) to get the CEP2 badge. I find something strange. In linux I finished a WU like c4cw and hcc in about 30-40% less time (which I hear is normal), but then I get rewarded with only 60% of the pts per hr. Is this normal? I mean the WUs are the same, but if you do it in linux you not only get less because you finish it faster, but you also get less per hour?
---------------------------------------- |
||
|
|
Dataman
Ace Cruncher Joined: Nov 16, 2004 Post Count: 4865 Status: Offline Project Badges: 
|
I just converted one of my pcs over to Linux (dual boot) to get the CEP2 badge. I find something strange. In linux I finished a WU like c4cw and hcc in about 30-40% less time (which I hear is normal), but then I get rewarded with only 60% of the pts per hr. Is this normal? I mean the WUs are the same, but if you do it in linux you not only get less because you finish it faster, but you also get less per hour? Yes this is normal. Credit Granting Rules These are the standard rules for granting Credit, based on the number of Results returned. Each Project will decide which rule is appropriate for their Project, or if they need to rely on a different method . Number of Results Rule 1 Result: The Claimed Credit is granted provided the Result passes some Project specific test. For example, Climateprediction.net (CPDN) grants a fixed amount for each "Trickle" reported. 2 Results: The lowest of the Claimed Credit values is granted. 3 Results: The top and the bottom of the Claimed Credit values are removed and the middle value is granted. Note: This is really just a special case of the 4 or more rule. 4 or more Results: The top and the bottom of the Claimed Credit values are removed and the middle values are averaged and the average of these Claimed Credits is granted. The difference of course is the "project specific test" which can be anything.   |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
2) At WCG is by exception the average and not the lowest in quorum 2, except with outliers where it looks at the history claim/grant and the one nearest is declared canonical. The system was supposed to check persistent underclaimers and normalize... but what if an overclaimer meets an underclaimer? After much experimenting found that http://boinc.berkeley.edu/dl/boinc_6.10.58_x86_64-pc-linux-gnu.sh is the version to go with on Linux 64 because it outputs the Windows benchmarks close enough that the claims approach sufficiently to end up with the average as grant instead of the low claim being declared canonical... makes a few hundred RAC difference. The only thing I did was:
----------------------------------------a) Install the 6.10.58 from the GetDeb ppa b) Dropped in the libraries and executables in place from the .sh... no actual install. c) Made sure the ownership and permissions were set identical and ran with it producing the windows equivalent benchmarks and quorum 1... is at the grace of the mini reference task benchmark excution inside the main task. Think for C4CW it's different to the AutoDock version as the execution times seem to be quite stable and thus knowing a large population credit set from Beta might be the base [wild guess]. I have them run in ranges of 1.41-1.48 hours and they barely vary a few tenth of a credit. Sample set, where the step change in claim signifies the drop in of the .sh kit... claim less, get more. c4cw_ target01_ 487462682_ 0-- 1292373 Valid 28-8-10 15:34:19 30-8-10 19:29:40 1.42 25.6 / 26.7 c4cw_ target01_ 487226532_ 0-- 1292373 Valid 28-8-10 15:34:00 30-8-10 19:00:56 1.41 25.4 / 26.8 c4cw_ target01_ 487377715_ 0-- 1292373 Valid 28-8-10 15:34:00 30-8-10 18:45:02 1.41 25.5 / 27.0 c4cw_ target01_ 487652617_ 0-- 1292373 Valid 28-8-10 15:34:00 30-8-10 17:38:17 1.43 25.7 / 26.6 c4cw_ target01_ 481585027_ 0-- 1292373 Valid 28-8-10 14:10:45 30-8-10 17:32:04 1.41 25.4 / 26.6 c4cw_ target01_ 481268306_ 0-- 1292373 Valid 28-8-10 14:02:24 30-8-10 17:23:22 1.41 25.4 / 26.5 c4cw_ target01_ 476152176_ 0-- 1292373 Valid 28-8-10 13:04:31 30-8-10 16:23:48 1.42 25.6 / 26.0 c4cw_ target01_ 475393324_ 0-- 1292373 Valid 28-8-10 12:44:58 30-8-10 16:23:48 1.42 25.6 / 26.0 c4cw_ target01_ 474681983_ 0-- 1292373 Valid 28-8-10 12:35:34 30-8-10 16:01:45 1.42 25.6 / 26.3 c4cw_ target01_ 474307804_ 0-- 1292373 Valid 28-8-10 12:32:48 30-8-10 15:55:53 1.42 25.6 / 26.3 c4cw_ target01_ 474451885_ 0-- 1292373 Valid 28-8-10 12:30:55 30-8-10 14:47:02 1.42 33.7 / 25.9 c4cw_ target01_ 474823099_ 0-- 1292373 Valid 28-8-10 12:30:35 30-8-10 14:35:21 1.41 33.6 / 25.9 c4cw_ target01_ 474722715_ 0-- 1292373 Valid 28-8-10 12:30:13 30-8-10 14:27:58 1.42 33.6 / 25.9 c4cw_ target01_ 474936390_ 0-- 1292373 Valid 28-8-10 12:29:32 30-8-10 14:26:15 1.43 34.0 / 25.9 c4cw_ target01_ 471398689_ 0-- 1292373 Valid 28-8-10 11:41:42 30-8-10 13:21:20 1.42 35.9 / 26.1
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Dataman
Ace Cruncher Joined: Nov 16, 2004 Post Count: 4865 Status: Offline Project Badges:
|
Think for C4CW it's different to the AutoDock version as the execution times seem to be quite stable and thus knowing a large population credit set from Beta might be the base [wild guess]. I have them run in ranges of 1.41-1.48 hours and they barely vary a few tenth of a credit. Mine are quite close together on Windows also. Always with a few hundredths of an hour. Linux take less time than Windows. Also C4CW is about 16 credits/hour vs. 12 on most other WCG projects. |
||
|
|
|