Message boards : Number crunching : Credit system not fair
Author | Message |
---|---|
Mod.Zilla Volunteer moderator Send message Joined: 5 Sep 06 Posts: 423 Credit: 6 RAC: 0 |
The original post was removed due to swearing. Here is the rest of the post by: schatten1411 Sorry, but i can't understand that. An FX60 with 2800MHz will makes more points (or CPU-power) than my X2-6000 with 3000MHz. Result: over 600pts to my 160 per day !!!! Others will tell me more then 1000 PC's their own !!!! Rosetta Informational Moderator: Mod.Zilla |
Bird-Dog![]() Send message Joined: 15 Dec 05 Posts: 42 Credit: 116,749 RAC: 0 |
The benchmarks are way down on your No1 computer. Measured floating point speed 1063.4 million ops/sec Measured integer speed 962.82 million ops/sec Something wrong for an XP6000+ Also you`ve only 255mb RAM. |
[RKN] schatten1411 , Mitglied des Teams und des VEREINS Rechenkraft.net Send message Joined: 25 Apr 07 Posts: 12 Credit: 441,995 RAC: 0 |
The benchmarks are way down on your No1 computer. Rechner-1 was an test by virt. PC. The real machine is a AMD X2 6000+ dual core with 4GB DDR2 800 4-4-4-4-12 Corsair RAM. Measured floating point speed 2769.28 million ops/sec Measured integer speed 5488.87 million ops/sec And the result after a more then 18000sec projekt 10.x claimed / 4.x cranted credit. Why not points by projekt? To calculate a protein u need x steps. x steps = x points. |
![]() Send message Joined: 3 Nov 05 Posts: 1833 Credit: 120,009,519 RAC: 6,828 ![]() |
it's difficult to help you when your computers are hidden as we can't see the benchmarks and the output from the results. Obviously a 3GHz Athlon should produce more points than a 2.8GHz Athlon per second of compute time (assuming both have 1MB cache per core), so there is something amiss on the 3GHz machine. It might be that jobs have been 'waiting for memory' and so only running on one core at a time - there are lots of possibilities, as is cool n quiet kicking in and reducing the clock speed. The benchmarks don't have too much effect on the credit granted, but they do indicate if there are problems. There are lots of people here that can help but the tone of your first post might put some off from helping you. |
Mod.Sense Volunteer moderator Send message Joined: 22 Aug 06 Posts: 4018 Credit: 0 RAC: 0 |
Why not points by projekt? To calculate a protein u need x steps. x steps = x points. Well, that might be a fair system if steps always took the same amount of effort to complete... but they don't. The Rosetta credit system is based upon the actual useful project work completed. If your machine was not able to complete as much useful work in 12hrs as others, perhaps you should consider asking about that, rather then assuming you understand all there is to know about the situation after 12hrs of observation. If others received more credit, perhaps they understand something you may not. Rosetta Moderator: Mod.Sense |
![]() Send message Joined: 21 Feb 07 Posts: 3 Credit: 1,500 RAC: 0 |
|
![]() Send message Joined: 3 Nov 05 Posts: 1833 Credit: 120,009,519 RAC: 6,828 ![]() |
It's strange... There is some natural variation in how much processing power it takes to complete a decoy. If you run more decoys on a given target (i.e. set the run-time to longer) then you should get less variation as the variation will average out over the different decoys. Returning a task soon after that Work Unit type has been released also leads to variation as the credits take a while to settle down (the points are based on a running-average of all the credit claims submitted so far). Other things play a role too - cache size, CPU architecture, FSB and memory speeds probably all affect different WUs and different hardware/software combinations differently. It averages out pretty well in the long run though. HTH Danny |
![]() ![]() Send message Joined: 30 Dec 05 Posts: 1755 Credit: 4,690,520 RAC: 0 |
Maion, the literal answer to your question is that others found the first task much easier to crunch then your machine did. And so the question becomes "why would my machine take longer on that first task?". In a nutshell, the benchmarks only measure CPU processing performance and none of the rest of what a computer must do to complete work (memory/bus performance, disk access time, number of L2 cache misses etc.). My guess is that your processor has only a small L2 cache, and that the first task happened to require more cache in order to run well. So, other machines that have more L2 cache didn't notice any difference between the two, and they were able to crunch the first just about as quickly as the last. Add this signature to your EMail: Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might! https://boinc.bakerlab.org/rosetta/ |
Beezlebub![]() Send message Joined: 18 Oct 05 Posts: 40 Credit: 260,375 RAC: 0 |
It's strange... Depends on type of workunit, a couple of mine are below: 97500392 88333567 5 Aug 2007 19:51:20 UTC 6 Aug 2007 11:06:31 UTC Over Success Done 28,828.25 126.40 145.84 97498718 88331995 5 Aug 2007 19:38:36 UTC 6 Aug 2007 8:29:21 UTC Over Success Done 28,694.06 55.13 67.92 97468695 88303891 5 Aug 2007 16:11:32 UTC 6 Aug 2007 3:05:14 UTC Over Success Done 28,822.30 126.38 140.54 Intel Q6600 2.5ghz 4meg cache 2 gig ram |
[RKN] schatten1411 , Mitglied des Teams und des VEREINS Rechenkraft.net Send message Joined: 25 Apr 07 Posts: 12 Credit: 441,995 RAC: 0 |
104486638 94815520 10 Sep 2007 1:49:17 UTC 11 Sep 2007 4:20:16 UTC Over Success Done 10,716.14 46.88 50.68 104486637 94815519 10 Sep 2007 1:49:17 UTC 11 Sep 2007 4:20:16 UTC Over Success Done 10,795.95 47.23 20.00 ?????????????? My new Quad Core PC with 2GB RAM has calculate this. Same CPU-time (into the same real time) but 40% granted credit for WU 104486637. I'm confused. Works the one core faster than the others ?? Or this one slower ?? Now this PC is SHUT DOWN for rosetta. THIS point system isn't fair and don't will be. |
![]() ![]() Send message Joined: 16 Jul 07 Posts: 132 Credit: 98,025 RAC: 0 |
104486638 94815520 10 Sep 2007 1:49:17 UTC 11 Sep 2007 4:20:16 UTC Over Success Done 10,716.14 46.88 50.68 I believe you need to post this on the "problems with 5.78" thread. There was an error on the 20.00 result Jmarks |
[RKN] schatten1411 , Mitglied des Teams und des VEREINS Rechenkraft.net Send message Joined: 25 Apr 07 Posts: 12 Credit: 441,995 RAC: 0 |
104486638 94815520 10 Sep 2007 1:49:17 UTC 11 Sep 2007 4:20:16 UTC Over Success Done 10,716.14 46.88 50.68 Too !! But i've more examples. Look this. 104323559 94707693 9 Sep 2007 15:24:06 UTC 10 Sep 2007 13:39:51 UTC Over Success Done 10,562.89 46.21 64.62 104323558 94707692 9 Sep 2007 15:24:06 UTC 10 Sep 2007 13:39:51 UTC Over Success Done 9,986.70 43.69 41.81 104323557 94707691 9 Sep 2007 15:24:06 UTC 10 Sep 2007 13:39:51 UTC Over Success Done 10,086.28 44.13 34.21 104323556 94707690 9 Sep 2007 15:24:06 UTC 10 Sep 2007 1:49:17 UTC Over Success Done 10,668.45 46.67 34.25 Nearly same CPU-time but granted credit from 34.25 to 64.62. It's mysterious. |
FluffyChicken![]() Send message Joined: 1 Nov 05 Posts: 1260 Credit: 369,635 RAC: 0 |
104486638 94815520 10 Sep 2007 1:49:17 UTC 11 Sep 2007 4:20:16 UTC Over Success Done 10,716.14 46.88 50.68 It is the average over time that Rosetta@Home aim for. But if you do not want to contribute to Rosetta@home because of some variation in credits between tasks, while can be quite large there are plenty of other projects like Seti@Home that have a much simpler program and so it is much simpler to give even credit. Just don't forget to use Seti@Home optimised programs like lunatics/kwsn programs. Or go and claim what you like at the one or two projects that still let you. It is up to you after all and your priorities. Team mauisun.org |
[RKN] schatten1411 , Mitglied des Teams und des VEREINS Rechenkraft.net Send message Joined: 25 Apr 07 Posts: 12 Credit: 441,995 RAC: 0 |
But the problem is standing. I`ve say: u need x steps to calculate a protein, more steps -> more points. U can receive this. Now credit will granted by "monday + apple * dogs u`ve seen at last week" and all in association to rotation of sun. It will don`t understand. No one can follow this pointsystem. CPU time 10.000, claimed 50, granted -> lets surprise. GREAT !!! |
Kyle Kopid Send message Joined: 22 Oct 06 Posts: 18 Credit: 123,521 RAC: 0 |
What are you gonna do with the credits when you get them? Does it really matter how many you get? |
Mod.Sense Volunteer moderator Send message Joined: 22 Aug 06 Posts: 4018 Credit: 0 RAC: 0 |
Please let's not start that. schatten has a point and I don't think there is anyone that disagrees with their point, that the current credit system has a much broader variance, especially between various type of tasks, then would be preferred. And if they feel that is not in line with their objectives, they are free to react as they wish. schatten thanks for posting your reasons for why you are leaving. At least that feedback can be helpful for future consideration. Rosetta Moderator: Mod.Sense |
Kyle Kopid Send message Joined: 22 Oct 06 Posts: 18 Credit: 123,521 RAC: 0 |
Ha, I'm just playing devil's advocate. I can see where the original poster is coming from. It would be nice to be involved in a project that gives more credits than average. |
![]() ![]() Send message Joined: 17 Sep 05 Posts: 211 Credit: 4,246,150 RAC: 0 |
The thing I think needs to be done to improve the credit system here is a longer baseline. Currently the first few tasks from a new set can get very skewed credit from the later tasks from that same set. I do not know if this would be practical or how many tasks would comprise a good baseline. BOINC WIKI ![]() ![]() BOINCing since 2002/12/8 |
Nothing But Idle Time Send message Joined: 28 Sep 05 Posts: 209 Credit: 139,545 RAC: 0 |
The thing I think needs to be done to improve the credit system here is a longer baseline. Currently the first few tasks from a new set can get very skewed credit from the later tasks from that same set. I do not know if this would be practical or how many tasks would comprise a good baseline. A personal observation: Some time ago I used to get a "premium" on my credit; that is, credit granted was usually at or above the claim. But after installing a newer version of boinc with "more accurate" benchmarking my granted is now usually BELOW the claim (not always). So if the benchmarking is supposed to be more in-line with my computer's actual performance, then it is a revelation to me to see that my granted is consistently below the claim, which is what I always suspected here at Rosetta. Secondly, I agree that a longer baseline or better way of averaging the returned results (baseline) might help. Every day more duos and quads are added that perhaps skew the baseline as well as which architectures run 3-hr versus 24-hr WUs and report first. There ought to be a way to mitigate both the architecture and the reporting period on the overall credit granted. OTH, if it delays the granting of credit then I suppose that will cause furor also. |
Mod.Sense Volunteer moderator Send message Joined: 22 Aug 06 Posts: 4018 Credit: 0 RAC: 0 |
Idle, the real question about your granted credit is... is it about the same prior to the BOINC change as it is after? If the credit system is working well, then the actual credit granted shouldn't have moved at all due to the changed benchmarks. Keck, I think it is more correct to say that none of us has any way to know if our skewed task was reported early or late, but we all PRESUME that those with radically different credit must have been early. Before very many results had gone in to the average. However, the number of reports of skewed credits seem to exceed the likelihood of that being the case. One idea, I'll just throw out here. If you take a longer target runtime, and/or increase your work cache, even just by a few hours, then you'll never be in the early reports back, and YOUR results should always be reported after a firmer baseline has been established. So, while I agree that approaches to getting a better baseline should be investigated and are desireable, you can accomplish that effect your self if you wish to do so. Rosetta Moderator: Mod.Sense |
Message boards :
Number crunching :
Credit system not fair
©2025 University of Washington
https://www.bakerlab.org