BURP credit system explained?

Message boards : Number crunching : BURP credit system explained?
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Profile mpan3

Send message
Joined: 24 Apr 06
Posts: 64
Credit: 29,899
RAC: 0
Message 4682 - Posted: 27 Feb 2007, 1:11:13 UTC

I am aware of the general BIONC way of handing out credit, but I can\'t seems to find a detailed explaination of how BURP hands out redit...
ID: 4682 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Energiequant
Project donor

Send message
Joined: 14 Mar 05
Posts: 89
Credit: 958
RAC: 0
Message 4697 - Posted: 27 Feb 2007, 12:03:57 UTC

I never noticed BURP to do it different than other projects? (quorum but still no validator (result comparism) as far as I know, average claimed credits are granted)
ID: 4697 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mpan3

Send message
Joined: 24 Apr 06
Posts: 64
Credit: 29,899
RAC: 0
Message 4701 - Posted: 27 Feb 2007, 18:32:18 UTC - in response to Message 4697.  

I never noticed BURP to do it different than other projects? (quorum but still no validator (result comparism) as far as I know, average claimed credits are granted)


i thought Janus put in the image validator already? since now every WU is send to at least two client...

Anyways, so bascially it\'s the (cpu_benchmark * cpu_time / some_constant) ? I was wondering because my client consistantly claim the least credit for the same WU out of the two/three clients.
ID: 4701 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
B-Roy

Send message
Joined: 5 May 05
Posts: 8
Credit: 152
RAC: 0
Message 4721 - Posted: 28 Feb 2007, 8:46:26 UTC

I finished one wu and find the credit system to be weird.
While I was claiming more than 70 credit (over 9 hours), another user claimed only 3.50 (900 sec), and we received 3.50, with a huge difference in used time.

See: http://burp.boinc.dk/workunit.php?wuid=111748 .
ID: 4721 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile RDC

Send message
Joined: 25 Mar 05
Posts: 18
Credit: 4,117
RAC: 0
Message 4747 - Posted: 1 Mar 2007, 21:49:11 UTC

I\'ve noticed that BURP seems to have much more wild fluctuation in credits on the same WU than other BOINC projects do. As a slower machine, I tend to claim more credit on BURP since it takes more CPU time to process but I almost always end up getting more chopped off the claimed credit when granted compared to other projects (except those that have changed to a more of a fixed style credit system like Einstein).

The lowest of 2 results used here really should be changed to a consensus of 3. It evens things out a bit more in my opinion.




ID: 4747 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile RDC

Send message
Joined: 25 Mar 05
Posts: 18
Credit: 4,117
RAC: 0
Message 4749 - Posted: 2 Mar 2007, 1:34:57 UTC

This is a good example:

http://burp.boinc.dk/workunit.php?wuid=120488

Unfortunately it took my CPU 13,266.31 to process and it claimed 21.21 credits. The other result turned in at this point took 611.17 seconds and claimed 2.33 so 2.33 will be awarded.

That\'s a significant difference in credits and time to process. It\'s also why I don\'t do very much work for BURP anymore since the more intensive the work to be rendered, the higher percentage of credit difference.

In all seriousness, the way credits are awarded currently should be looked at before coming out of Alpha. Right now, it probably is driving more people off than it\'s attracting.
ID: 4749 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mpan3

Send message
Joined: 24 Apr 06
Posts: 64
Credit: 29,899
RAC: 0
Message 4751 - Posted: 2 Mar 2007, 5:01:41 UTC - in response to Message 4749.  

RDC, the \'other\' system is a core 2 duo, which is many times faster than your athlon, which could explain why it crunched through the WU so fast and claimed so little credit. If BURP does uses the original BIONC credit system (janus, can you please confirm this?) then i think there won\'t be any change to the credit system.

From what I understand, BIONC uses the cpu-clock to time how long a WU takes, not the wall-clock. So in theory, it is more accurate because it reflects how long your system actually spent on the WU, not how much time has passed since the system starts to work on the WU. (the difference is obvious only when BURP is not taking up 100% of the CPU), it then multiplies this time value by your performance benchmark score, thus effectively calculated how much work was really done on the WU, finally the client then claims a specific amount of credit proportional to that value.

So, if your client is claiming too much credit compare to everyone else, either your benchmark score is too high, or that your system somehow spent much longer to render the scene (maybe not enough RAM and thus HDD thrashing?)
Contribution to BURP total: 0.5% (manually updated)
ID: 4751 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Professor Desty Nova
Avatar

Send message
Joined: 21 Mar 05
Posts: 97
Credit: 291,318
RAC: 0
Message 4757 - Posted: 2 Mar 2007, 9:10:21 UTC

If you see on the front page, the credit system still needs lots of work. Right now it uses the default BOINC system. But the major problem right now is that the client (the blender module) is miscalculating the time. Instead of counting CPU time, is counting Wall time most of the time, and sometimes it gives lower results.


Professor Desty Nova
Researching Karma the Hard way
ID: 4757 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Janus
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 16 Jun 04
Posts: 4563
Credit: 2,097,282
RAC: 0
Message 4760 - Posted: 2 Mar 2007, 11:08:34 UTC - in response to Message 4757.  

If you see on the front page, the credit system still needs lots of work. Right now it uses the default BOINC system. But the major problem right now is that the client (the blender module) is miscalculating the time. Instead of counting CPU time, is counting Wall time most of the time, and sometimes it gives lower results.

Exactly, the relevant link is to this thread.
ID: 4760 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
PovAddict
Avatar

Send message
Joined: 25 Apr 05
Posts: 347
Credit: 4,618
RAC: 0
Message 4834 - Posted: 5 Mar 2007, 3:19:28 UTC
Last modified: 5 Mar 2007, 3:20:36 UTC

Credit should, in theory, be the same for all machines. A slower machine takes more time, but also has lower benchmarks. In theory, they should be proportional, in the sense that a half-speed computer should have half the benchmarks and twice the crunching times, getting the exact same credits.

In theory, practice and theory are the same. In practice, they are not.

The problem on BURP is the CPU time being miscalculated, that\'s it.
ID: 4834 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
B-Roy

Send message
Joined: 5 May 05
Posts: 8
Credit: 152
RAC: 0
Message 4850 - Posted: 5 Mar 2007, 15:24:01 UTC

i think the problem is that there is a quorum of 2 and the lower claimed credit is granted, which can\'t be the way it is going to stay. Either there will be a switch to a quorum of 3 + the canonical result counts, or what I prefer fixed credit for the amount of work done.
ID: 4850 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile ashriel

Send message
Joined: 2 Mar 07
Posts: 1
Credit: 115
RAC: 0
Message 5066 - Posted: 23 Mar 2007, 21:41:00 UTC
Last modified: 23 Mar 2007, 21:42:17 UTC

Hmmmm..
I always get much lower credits then claimed - between 30 and 85% of claimed to be exactly:
http://burp.boinc.dk/results.php?hostid=26632
It\'s a lil bit disappointing.
ID: 5066 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
axarydax

Send message
Joined: 8 Mar 07
Posts: 11
Credit: 1,604
RAC: 0
Message 5143 - Posted: 28 Mar 2007, 13:32:40 UTC

you have only athlons, maybe the other computers who do the same work are some faaaast core2 boxes, who claim less credit...

at least with my core2 I get usually as much credit as claimed.
ID: 5143 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mikie Tim T

Send message
Joined: 21 Apr 05
Posts: 16
Credit: 57,866
RAC: 0
Message 6232 - Posted: 28 Jun 2007, 13:28:43 UTC

Is the issue with the miscalculated CPU time ironed out yet? Or is this still an issue with the current version?

ID: 6232 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
zombie67 [MM]
Project donor
Avatar

Send message
Joined: 9 Dec 06
Posts: 91
Credit: 2,118,668
RAC: 6
Message 6254 - Posted: 30 Jun 2007, 6:23:29 UTC

It\'s still completely screwed.
Reno, NV
Team: SETI.USA

ID: 6254 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Janus
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 16 Jun 04
Posts: 4563
Credit: 2,097,282
RAC: 0
Message 6267 - Posted: 1 Jul 2007, 9:02:41 UTC - in response to Message 6254.  

It\'s still completely screwed.

In what way exactly?
ID: 6267 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
zombie67 [MM]
Project donor
Avatar

Send message
Joined: 9 Dec 06
Posts: 91
Credit: 2,118,668
RAC: 6
Message 6276 - Posted: 1 Jul 2007, 21:30:33 UTC - in response to Message 6267.  

It\'s still completely screwed.

In what way exactly?

http://burp.boinc.dk/workunit.php?wuid=423694

o My machine (a 2x Clovertown Xeon X5355 @ 2.66ghz) took 107,971 seconds and claimed 534. FP=2620.51. Int=5817.91.

o Another machine (PD @ 3.0ghz) took 30,962 seconds and claimed 62. FP=1372.94. Int=2139.09

o The third machine (C2D @ 2.4ghz) took 28,149 seconds and claimed 107.90. FP=2058. Int=4549.91


Now, these other two machines may be OC. But there is *no way* they are 3-4 times faster than mine. And the benchmarks don\'t show much, if any, OC.

Something is really wrong with the way time is calculated. Or maybe there is some bizarre variation, where the same result crunched on the same machine multiple times, comes up with different (real) run times...if that\'s possible.

In any case, I was awarded significantly less credits than I should have been.


Reno, NV
Team: SETI.USA

ID: 6276 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keck_Komputers
Avatar

Send message
Joined: 6 Mar 05
Posts: 94
Credit: 1,378,284
RAC: 0
Message 6277 - Posted: 2 Jul 2007, 0:39:57 UTC

It looks like the host mentioned is seriously overclaiming. All of the tasks I looked at yours was the highest claim. Rerunning benchmarks may help. When the 5.10.x client is released it may help as well there is another attempt to fix the benchmarks in that version. The problem is that benchmarks do not always acurately predict what the real world performance of a computer will be.

Because of this benchmark problem many BOINC projects have changed to a different way of granting credit. I do not think that will be an option at this project. The complexity of the tasks is unknown and can not be predicted accurately in any way I can think of. So the only way may be to actually crunch the task and see what we get.
BOINC WIKI

BOINCing since 2002/12/8
ID: 6277 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
zombie67 [MM]
Project donor
Avatar

Send message
Joined: 9 Dec 06
Posts: 91
Credit: 2,118,668
RAC: 6
Message 6278 - Posted: 2 Jul 2007, 3:02:04 UTC - in response to Message 6277.  
Last modified: 2 Jul 2007, 3:03:31 UTC

It looks like the host mentioned is seriously overclaiming. All of the tasks I looked at yours was the highest claim. Rerunning benchmarks may help.



Do you mean my machine? Look at the benchmarks. They are not out of line. Compare to the slower C2D on the same WU.

Claims are based on the benchmarks, and the benchmarks are regular/normal.

Look at the TIMES. Are you saying it is normal for my machine to take 3-4 times longer? Even though it is a faster machine? Something is wrong with the times.

Reno, NV
Team: SETI.USA

ID: 6278 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
zombie67 [MM]
Project donor
Avatar

Send message
Joined: 9 Dec 06
Posts: 91
Credit: 2,118,668
RAC: 6
Message 6279 - Posted: 2 Jul 2007, 3:20:20 UTC

Here is another, with a different computer of mine.

http://burp.boinc.dk//workunit.php?wuid=421022

o PD @ 2.66. 54,464. seconds. Claimed 115. FP=1431.88. Int=2385.02.

o C2Q @ 2.66, OC to 3.30. 27,654 seconds. Claimed 164. FP=3232. Int=7245.86. (this is mine)

o C2D @ 2.40. 19,743 seconds. Claimed 78. FP=2217.33. Int=4628.44.

So how is it the C2D beat my C2Q by *29%*? *Maybe* the C2D is OC *far* beyond my C2Q, but then the benchmarks have been artificially adjusted down. Not likely.

I think the benchmarks are all correct, and the claims based on them are correct. What are *not* correct in all of this are the crunch times.
Reno, NV
Team: SETI.USA

ID: 6279 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
1 · 2 · Next

Message boards : Number crunching : BURP credit system explained?