| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: BOINC Points post in Member News |
| No member browsing this thread |
|
Thread Status: Locked Total posts in this thread: 277
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
knreed,
Strongly in favour of project benchmark wu's, especially if they are 'requestable' - a great facility for those who 'tune' their systems. I for one would far rather measure my system performance in terms of hdc std wu in x;:x hrs:mins, rather than super pi.... I do think however your approach could be simplified a little. Step 1 is perhaps unnecessary, as you are simply trying to derive an average value from which variance, and therefore credit, for users systems is measured. It would be valid to arbitarily set this 'average' based on your own bechmarks. I still believe it would be better, that, given your 'computed average', you use this as a yardtick to measure the 'size' of wu's that go out, and award credit relative to the size/average. In this scenario, no client benchmarking is required - all we need to do is to turn those wu's around faster to get more credit, and in the process, do more science... The only drawback of this approach i can forsee is when a project has wildly varying wu's. I believe if your workstreams are sampled enough then an average value can be derived for the stream as a whole. It's simple, and ultimately fair to all. We will then be able to move discussions away from frigging credit / benchmarks to the cheek of those cherry-picking short wu's  Jonathan |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The problem with any PC specific benchmark is it may not always be the same at any given moment, the same as it is now. It is dependent on resources allocated at that present time, which 30 minutes later maybe different. We see this problem now with the benchmarking.
----------------------------------------I think it has to be the WU that is benched not the machine, at present I also feel that people are trying to be too accurate in these measures. Take an faah WU, on my C2D it takes between 3.2 - 3.7 hours on my 3800+`s 6-7hours........I`d be happy to take a mean 60 points for each of those knowing everyone else would get the same. I think precision is what is causing the problems and an average would actually be better and more acceptable. Yes there would be times when the odd wu differed but it would even itself out across all ! [Edit 1 times, last edit by Former Member at Nov 14, 2006 11:41:36 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I still have problems with a D930 @3600Mhz getting low benches and lowering the median.
Below is recent scores for this Workunit Name Device Name Status Sent Time Time Due / Return Time CPU Time (hours) Claimed/ Granted BOINC Credit faah0893_ bdb810_ mx8hvp_ 06 psycho-240756 Valid 11/13/2006 16:59:21 11/14/2006 12:45:01 4.38 44 / 53 faah0893_ bdb033_ mx7upj_ dry_ 05 psycho-240756 Valid 11/12/2006 17:05:45 11/13/2006 07:44:56 4.30 41 / 60 faah0892_ bdb820_ mx7hvp_ dry_ 00 psycho-240756 Valid 11/12/2006 14:51:23 11/13/2006 07:44:56 4.65 45 / 59 faah0892_ bdb700_ mx7upj_ 01 psycho-240756 Valid 11/12/2006 12:21:54 11/12/2006 21:38:45 4.80 46 / 59 faah0892_ bdb649_ mx7upj_ 00 psycho-240756 Valid 11/12/2006 10:22:53 11/12/2006 19:32:04 4.71 45 / 54 faah0892_ bdb254_ mx7hvp_ dry_ 00 psycho-240756 Valid 11/11/2006 21:25:23 11/12/2006 14:51:23 4.56 43 / 56 faah0892_ bdb113_ mx7upj_ 03 psycho-240756 Valid 11/11/2006 17:10:08 11/12/2006 10:22:53 4.62 44 / 52 faah0892_ bdb023_ mx7upj_ 04 psycho-240756 Valid 11/11/2006 14:18:04 11/11/2006 21:25:23 4.27 41 / 50 faah0891_ bdb370_ mx7hvp_ 02 psycho-240756 Valid 11/10/2006 23:06:39 11/11/2006 17:10:08 4.59 44 / 50 faah0891_ bdb324_ mx5hvp_ dry_ 09 psycho-240756 Valid 11/10/2006 21:26:17 11/11/2006 17:10:08 4.78 45 / 50 faah0891_ bdb304_ mx7hvp_ 01 psycho-240756 Valid 11/10/2006 20:16:45 11/11/2006 12:33:18 4.94 47 / 51 You will note it is benching low and therefore pulling average down to it`s level, compare this with my C2D @3300Mhz Workunit Name Device Name Status Sent Time Time Due / Return Time CPU Time (hours) Claimed/ Granted BOINC Credit faah0893_ bdb856_ mx7upj_ dry_ 04 p5ycho-77f4c849 Valid 11/13/2006 18:09:44 11/14/2006 11:11:26 3.49 69 / 71 faah0893_ bdb775_ mx7upj_ dry_ 04 p5ycho-77f4c849 Valid 11/13/2006 15:50:45 11/13/2006 23:06:45 3.46 68 / 77 faah0893_ bdb745_ mx7upj_ dry_ 01 p5ycho-77f4c849 Valid 11/13/2006 15:06:08 11/13/2006 21:09:41 3.48 69 / 66 faah0893_ bdb158_ mx7upj_ dry_ 01 p5ycho-77f4c849 Valid 11/12/2006 21:16:04 11/13/2006 12:16:45 3.31 66 / 72 faah0893_ bdb116_ mx8hvp_ 05 p5ycho-77f4c849 Valid 11/12/2006 19:38:03 11/13/2006 09:33:47 3.14 62 / 63 faah0893_ bdb055_ mx7upj_ dry_ 03 p5ycho-77f4c849 Valid 11/12/2006 17:58:51 11/13/2006 09:33:47 3.03 60 / 56 faah0892_ bdb851_ mx7upj_ 04 p5ycho-77f4c849 Valid 11/12/2006 15:47:18 11/12/2006 22:35:27 3.15 63 / 69 faah0892_ bdb798_ mx7hvp_ dry_ 01 p5ycho-77f4c849 Valid 11/12/2006 14:08:08 11/12/2006 22:35:27 3.25 65 / 63 faah0892_ bdb741_ mx7hvp_ dry_ 07 p5ycho-77f4c849 Valid Redoing the benches on the C2D two day`s ago bought up the score somewhat from a similar situation howether I tried this on the D930 at the same time, no difference at all. It is pulling down the median I feel and generally scoring 10+ points less per WU |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
For what its worth, the D930 seems to claim low, but gets some compensation thru the higher claims of other quorum members.
----------------------------------------The C2D is well withing range. A FAAH however fast or slow crunched is worth 65 +/- 5 percentage points up or down. On the whole u seem to be in the ballpark with that one. What OSses are these babies on..... u may have stated it, but dont see it immediately. cheers
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sek they`re all on XP, seem`s silly to run linux howether quick it is when the points are extremely low with system.
I would like to see how fast the C2D would be on a linux system. Yes the D930 get`s a little compensation but IS IT pulling the whole quorum down so others suffer for my low claim ? |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Carl.h,
----------------------------------------it's probabilities. One has to see the detail quorum 3 records for each of the samples to determine if it was pulling the median down below to what the old middleman was. If u had a quorum of 45, 48, 58, one gets e.g. 50, which is 2 above what would be given under the old rule. In effect, 6 more credits in total are handed out over the old system. Of course no-one denies it could go the other way, but that was equated into the algorithms for a larger population. cheers
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
If you look at the results for the D930 it is consistantly low so I can only presume it is pulling the total points of a quorum down.
In effect you`ve stated as we`ve all seen that good machines should pull in 60 + points per faah, the D930 is well below consistantly it`s doubtful it`s fell everytime to others in a low bracket, one must presume it is pulling others down too ! |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
More info to come on this but this use of the average of the 3 members of the quorum vs the old way of using the middle now adds a greater "weight" when someone underclaims. Before the underclaimer was essentially ignored but now that same underclaimer effects the credits of all 3 parties in every WU that he/she participates in.
----------------------------------------I applaud what WCG was trying to do with the latest change but this part has had a very undesirable effect. There are also enough of these underclaimers out there that they are having the effect of pulling down the scoring and by that are skewing any competitions that are going on by having the net effect of reducing the overall points awarded. That is just a gut feeling based on observation so far but I'm having numbers put together to support that assumption. I'd really like to see this part of the change reversed.  [Edit 1 times, last edit by Movieman at Nov 15, 2006 1:28:28 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Did some basic averaging this morning over the last 10 faah wu`s on the D930 @3600 and a AMD 3800+ X2.
----------------------------------------Although the D930 appears to claim very low and does seem to pull the quorum down there is only a difference of 9 points over the two machines. This could be explained by one or two big scores from the D930 where it may have been eliminated by extremely low claim from the quorum eg., a 78 given where a 44 was claimed. I will keep my eye on this and in a week or howether long average over 100 faah to get a better look. At present it is not looking as bad as I expected but please reserve judgement. Over the last 10 average faah Wu`s (a couple were excluded for being too short). These are all dual machines. D930 X2 @ 3600 = 601 points AMD X2 3800+ @2100 = 610 points C2D E6600 @3300 = 647 points I have another dual 3800+ to include in bigger analysis when I get 100 WU`s on each. It`s certainly not the significant difference I expected as of yet. All machines use stock Boinc 5.4.11. [Edit 1 times, last edit by Former Member at Nov 15, 2006 6:12:57 PM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Carl.h
----------------------------------------The way i have from the beginning and from the supplemental posts by knreed understood is, that no-one but the 'outliers' would be negatively impacted. On a range of 10 that still could happen, but as u will collect more and more results under the new system, one should on the average get closer to the middle. The more 'opti's revert to stock and they will be driven to it, the less a quorum will be hit by an excluded 'outlier', thus pulling up the average. Indeed we should reserve judgment until we have a substantial body of work. cheers
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|