| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Unfair amount of credit? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 42
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I feel that there is more of an issue here than most realise. First, just to clarify a misconception, my experience has been, that it is the slow machines which claim the most points, not the fastest. From my experience, this is the opposite. BOINC did penalise the fastest machine within the quorum. It is difficult to get a true idea of what is and isn't happening over all with small data samples. Here is why I said what I did from my own recent claims (not points awarded): P111@598- mid 90's to mid 100's / FAAH project P111@998- low 90's to low 100's " " " Cely@1300-mid 80's to mid 90's " " " Cely@3600-Hi 70's to hi 80's " " " (overclocked from 2400) AMD@2100 low 60's to low 70"s " " " (XP 3000+ Barton Core) AMD@2200 Low 60's to high 60's " " " (XP 3200+ Barton Core) P4 @3200 Low 60's to mid 60's " " " (Hyperthreaded Prescott) As this is all the data to hand the trend line seems obvious. Clearly more data would be most useful. Cheers. ozylynx  |
||
|
|
Electrolyte
Cruncher Joined: Jun 20, 2006 Post Count: 13 Status: Offline |
Electro. It would seem that the Linux is the governing factor on your machines points. I only have one machine set up for linux but it drops to only 55% of the points claimed on Windows. Productivity is about the same but a different points calculation benchmarking is applied to Linux. cheers. ozylynx Now this has put me off projects that rely on the benchmarks. Any idea why the benchmarks are so low on Linux? Even an optimised BOINC client doesn't get as high benchmarks as Windows on a standard client. This isn't really fair on us Linux users   |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi again
Atm there is no answer to this problem with Linux benchmarking. It is being "looked into" and solutions are at hand  This project is mooted to be going to fixed points score system in the future but there is no forecast on that right now. We all live in hope. Cheers. ozylynx |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
BOINC uses exactly the same benchmark on Windows and Linux (unless you are cheating, of course).
Awarding fair points across different projects is one of the stated aims of any BOINC points shakeup. Also, equal points for equal work. The current system works reasonably well if nobody cheats, but it is pitifully easy to take advantage of. And, obviously, it only manages to be reasonably fair on average - individual granted credits can deviate quite wildly. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Please show me how it is fair again, I must have missed that bit.
Awarding fair points across different projects is one of the stated aims of any BOINC points shakeup. Also, equal points for equal work. I'm sorry but they missed their aim on all stated goals. BOINC uses exactly the same benchmark on Windows and Linux (unless you are cheating, of course). This is purely pedantic. Yes it is the same benchmark but it is unfair to Linux users. Nothing new here, we've discussed this before. The current system works reasonably well if nobody cheats, but it is pitifully easy to take advantage of. And, obviously, it only manages to be reasonably fair on average - individual granted credits can deviate quite wildly. I don't get it. The system doesn't work at all because it is so open to cheating from both sides of the fence. Linux boxen are teamed with other Linux boxen by default. Read my previous post about average consistant points deviation between CPU types and speeds. If Linux boxen can be grouped so can CPU types. I'm not saying that they are but it is clearly open to that. One could look after ones mates couldn't one? The fact that the CPU type produces benchmark results which can "deviate quite wildly" should be enough to have a fire alarm ringing in any thinking persons head. How is it even a remote possibility to have a "fair" system based on an unfair benchmark? Face facts..The entire system is and has been critically flawed since its inception. I await the new system patiently but sorry, I cannot stabd by and have the existing one called fair. Cheers. ozylynx |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Perhaps I wasn't clear, since you have misinterpreted just about everything I said.
As you say, this is an old discussion. But I must insist: any Linux/Windows variation exists only in your mind. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
OK Here's what my mind is going to do:
When the current WU finishes on the Celeron 1300, about 90 minutes from now, I will unplug the HDD with Windows XP on it and plug in the HDD with Linux on it. As this machine takes roughly 12.5 hours to complete a WU. I will report the claimed points for only 3. I feel certain that the difference will by then be obvious. No other changes will be made to this computer. Here are the "claimed points" for my most recent work on it: 86,84,83,81,85,97,82,95,84,80,95,85,96,81. I will post the Linux claim in a couple of days. Cheers. ozylynx |
||
|
|
Electrolyte
Cruncher Joined: Jun 20, 2006 Post Count: 13 Status: Offline |
As this machine takes roughly 12.5 hours to complete a WU. I will report the claimed points for only 3. I feel certain that the difference will by then be obvious. No other changes will be made to this computer. Here are the "claimed points" for my most recent work on it: 86,84,83,81,85,97,82,95,84,80,95,85,96,81. Erm, wow? What are the benchmarks for that system? To add a comparison, here is what my machines get: FX-55, average 3 hours per FAAH WU (benchmarked approx 1350 double precision MIPS (Whetstone) per CPU and 2713 integer MIPS (Dhrystone) per CPU) 22,23,21,20,24 average claimed credit P4, average 2 FAAH WUs every 5 hours (due to HT) (benchmarked approx 640 double precision MIPS (Whetstone) per CPU and approx 1100 integer MIPS (Dhrystone) per CPU) 21,22,21,20,22 average claimed credit To add to that, check out this statistics page: http://www.boincsynergy.com/stats/teams.php?project=wcg&team=3699 Narwhal is the one with the one dual-Xeon @ 3GHZ running. If you click on my name you'll see the 2 machines running Linux. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I'm not very conversant with this type of benchmarking so my terminology may not be too accurate. Here's what I have from Boincview, which is the only place i know where to look.
Total FPU speed = 1.179 GFlops FPU Speed = 1.179 GFlops Total Integer Speed = 2.09 GIops Credits/Hour = 6.81 Win XP I haven't been able to get it working with BOINCView on Linux. I'm actually holding off learning and using Linux because of the points problem on Boinc. When that is fixed I will be converting to Linux wholesale. In the interim I'm locked into Windoze by WCG and BOINC. I also can't bring up my previous stats for this box under Linux here, that's why I need to repeat this excercise. For interest sake they are still on file, partially, at BAM Boincstats and the few days I ran it under Linux produced...43, 54, 58, 84, 85 Points actual awarded. I think the last two are 2 WU each, giving a 42 point average, but I can't prove it. As I recall I was claiming around 55 points per WU. Cheers. ozylynx |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Surely u guys know how to get to the 'Result Status' and filter on WU's 'Valid'. If not, take the link below, select the 'Device Name' under discussion and pick 'Valid' from the 'Result Status' drop down box and filter.
----------------------------------------https://secure.worldcommunitygrid.org/ms/viewBoincResults.do?filterDevice= Here's the first page dump...generally the credit is better or equal to claim, with an occasional dipper. Workunit Name Device Name Status Sent Time Time Due / Return Time CPU Time (hours) Claimed/ Granted BOINC Credit B04573_ 0011_ CTMA4B-38-1-1-c1 tpdc-mxirnto2ft Valid 10/10/2006 15:55:33 10/13/2006 08:12:58 1.95 16 / 16 B04570_ 0148_ CTMA4B-36-25-16-c2 tpdc-mxirnto2ft Valid 10/10/2006 12:17:01 10/13/2006 08:12:58 1.35 11 / 14 B04570_ 0146_ CTMA4B-36-25-10-c2 tpdc-mxirnto2ft Valid 10/10/2006 12:17:01 10/13/2006 03:19:08 2.02 16 / 14 B04566_ 0103_ CTMA4B-34-2-7-c2 tpdc-mxirnto2ft Valid 10/10/2006 05:43:49 10/11/2006 14:34:59 2.38 19 / 21 B04566_ 0104_ CTMA4B-34-2-8-c2 tpdc-mxirnto2ft Valid 10/10/2006 05:43:49 10/12/2006 21:44:23 2.20 18 / 23 B04562_ 0073_ CTMA4A-38-16-15-c2 tpdc-mxirnto2ft Valid 10/09/2006 21:41:57 10/11/2006 05:40:16 1.93 16 / 16 B04562_ 0067_ CTMA4A-38-15-24-c2 tpdc-mxirnto2ft Valid 10/09/2006 21:41:57 10/11/2006 05:40:16 1.93 16 / 16 B04559_ 0256_ CTMA4A-37-9-15-c1 tpdc-mxirnto2ft Valid 10/09/2006 17:00:03 10/12/2006 20:27:23 1.81 15 / 19 B04559_ 0147_ CTMA4A-37-21-4-c1 tpdc-mxirnto2ft Valid 10/09/2006 16:11:08 10/12/2006 15:28:28 2.24 18 / 25 B04556_ 0168_ CTMA4A-35-22-21-c2 tpdc-mxirnto2ft Valid 10/09/2006 09:52:26 10/12/2006 12:40:59 1.88 15 / 15 B04555_ 0164_ CTMA4A-35-22-15-c1 tpdc-mxirnto2ft Valid 10/09/2006 07:32:31 10/12/2006 10:23:42 1.93 16 / 20 B04555_ 0099_ CTMA4A-35-16-23-c1 tpdc-mxirnto2ft Valid 10/09/2006 07:02:22 10/12/2006 07:03:18 1.81 15 / 20 B04555_ 0100_ CTMA4A-35-16-24-c1 tpdc-mxirnto2ft Valid 10/09/2006 06:56:51 10/11/2006 08:48:49 1.83 15 / 25 This section retains all WU's up to 4 days after quorum-3 is established. U can calculate the hourly claim....about 8.1 consistently. As for the benchmark value, simply sum the Dhry+Whetstone MIPS. For each 5 points increase, the hourly claim goes up or down by 0.01 (on stock WOSXP/BOINC). Why BOINC produces a different benchmark hourly on Linux versus BOINC, don't know. Not seen an answer on the official BOINC forum. http://boinc.berkeley.edu/dev/forum_index.php To hear it from a BOINC developers mouth, posted following question for the Rom-World Q&A. Hot topic: Why is the hourly benchmark value between Linux and Windows different, or so it's claimed. When done with stock BOINC 5.4.9 e.g. on Windows it kicks out 8.1 per hour, when same done under Linux, it kicks out 5.0. The WU's are processed at equal speed i.e. a job on Windows taking 2 CPU hours would take near equal time on Linux. thx
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|