Message boards : Theory Application : strange discrepency in credit points for Theory
Message board moderation

To post messages, you must log in.

AuthorMessage
Erich56

Send message
Joined: 18 Dec 15
Posts: 1119
Credit: 20,611,396
RAC: 38,319
Message 38766 - Posted: 11 May 2019, 6:36:08 UTC

can anyone explain the following phenomenon:

---------------------- Total time --- CPU time --- credit points

task from May 9: 63,508.37--- 60,515.58-- 2,715.47
task from May10: 64,895.94-- 61,689.60----- 313.00

same PC, same setting, everything else the same
ID: 38766 · Report as offensive     Reply Quote
maeax

Send message
Joined: 2 May 07
Posts: 704
Credit: 26,315,595
RAC: 35,836
Message 38767 - Posted: 11 May 2019, 6:52:18 UTC - in response to Message 38766.  

Yes,
the Credits went to normal. About 500 Cobblestones per day. Look at WCG for example.
ID: 38767 · Report as offensive     Reply Quote
Erich56

Send message
Joined: 18 Dec 15
Posts: 1119
Credit: 20,611,396
RAC: 38,319
Message 38768 - Posted: 11 May 2019, 7:20:22 UTC - in response to Message 38767.  

so before they were "abnormal"? How come?
Looks like lottery
ID: 38768 · Report as offensive     Reply Quote
Profile Laurence
Project administrator
Project developer

Send message
Joined: 20 Jun 14
Posts: 268
Credit: 226,909
RAC: 0
Message 38790 - Posted: 13 May 2019, 8:30:36 UTC - in response to Message 38768.  

so before they were "abnormal"? How come?
Looks like lottery


Credit has always been an issue. I don't know why the Theory app was so wrong but we have switched to [url=https://boinc.berkeley.edu/trac/wiki/CreditOptions runtime credit.[/url]. I believe that this is the best approach. Wall time on the host is the most important component and peak flops is a good enough value for normalization. The efficiency of the application on that host or between applications is an opportunity cost and should not be related to credit.
ID: 38790 · Report as offensive     Reply Quote
Vortac

Send message
Joined: 22 Apr 19
Posts: 1
Credit: 37,470,405
RAC: 150,247
Message 38887 - Posted: 18 May 2019, 7:22:13 UTC - in response to Message 38790.  

Credit has always been an issue. I don't know why the Theory app was so wrong but we have switched to [url=https://boinc.berkeley.edu/trac/wiki/CreditOptions runtime credit.[/url]. I believe that this is the best approach. Wall time on the host is the most important component and peak flops is a good enough value for normalization. The efficiency of the application on that host or between applications is an opportunity cost and should not be related to credit.

Browsing the top cop computers, one can easily see that people there are still getting even 18k credits for single-threaded Theory tasks. Others are getting mere 500. Obviously, credits urgently need further tweaking because this is a ridiculous imbalance:

https://lhcathome.cern.ch/lhcathome/results.php?hostid=9995505&offset=0&show_names=0&state=4&appid=13
ID: 38887 · Report as offensive     Reply Quote
Profile HerveUAE
Avatar

Send message
Joined: 18 Dec 16
Posts: 123
Credit: 33,810,943
RAC: 71,629
Message 38900 - Posted: 19 May 2019, 17:34:53 UTC

I observed a sudden increase in the credits given to Theory jobs on the 09/05/2019. This increase only occurred on 2 of my 3 rigs.

The increase do not seem to accurately reflect CPU usage: if you look at Yeti's top RAC computers, you will find the same spec in the top 3.
Interestingly, the first one (doing mainly Theory jobs) has 4 times more RAC than the next 2 which are doing mainly ATLAS jobs.

See here: https://lhcathome.cern.ch/lhcathome/hosts_user.php?sort=expavg_credit&rev=0&show_all=0&userid=555
We are the product of random evolution.
ID: 38900 · Report as offensive     Reply Quote
Erich56

Send message
Joined: 18 Dec 15
Posts: 1119
Credit: 20,611,396
RAC: 38,319
Message 38901 - Posted: 19 May 2019, 18:02:10 UTC - in response to Message 38900.  

This increase only occurred on 2 of my 3 rigs. The increase do not seem to accurately reflect CPU usage
same thing here. The credit points are totally different between my rigs, not in line at all with the CPUs.
It' all a big conondrum. Would be great if finally someone from LHC could give us a logical explanation for these descrepencies.
ID: 38901 · Report as offensive     Reply Quote
MaBoinc

Send message
Joined: 23 Mar 18
Posts: 2
Credit: 29,421,664
RAC: 118,930
Message 38981 - Posted: 26 May 2019, 17:24:35 UTC
Last modified: 26 May 2019, 17:25:38 UTC

From what I can tell it seems to be due to the large differences in device peak flops (the one which you can find above every task). Like this task https://lhcathome.cern.ch/lhcathome/result.php?resultid=230370834 from Vortac has a value of 152.78 GFLOPS on a 2950x, while I'm running a 1950x at 3Ghz @±37degC with quad channel 3000c15 memory, yet somehow this task: https://lhcathome.cern.ch/lhcathome/result.php?resultid=230662993 gives a value of 3.42 GFLOPS. Sure he might not have underclocked his Threadripper and it could very well be running on faster memory, but this seems a bit too extreme. Mine also used to perform perfectly fine before the credit update.
I updated virtualbox to the same version he is running but that probably won't make a difference, also ran the whetstone benchmark again, but not sure that will help either. Will report back if anything changes. Hope it can be fixed to be slightly more reflective of actual device performance.
ID: 38981 · Report as offensive     Reply Quote
Profile MAGIC Quantum Mechanic
Avatar

Send message
Joined: 24 Oct 04
Posts: 839
Credit: 36,634,197
RAC: 35,396
Message 39502 - Posted: 6 Aug 2019, 22:44:46 UTC
Last modified: 6 Aug 2019, 22:48:13 UTC

I still see many computers running these with single and 2-core tasks that average 350 Credits but others also running only 1-core tasks getting 5,000 - 23,000 Credits running the same tasks and not with any better CPU's or ram so that isn't the reason

But when you have them running 24/7 getting 5,000-23,000 Credits compared to the 350 average there sure is a BIG RAC and Total difference here this year.
ID: 39502 · Report as offensive     Reply Quote
wHewitt

Send message
Joined: 19 May 14
Posts: 7
Credit: 77,524,845
RAC: 738,160
Message 39503 - Posted: 7 Aug 2019, 2:29:57 UTC

Since I started crunching this project again after a year or so break...my RAC has increased significantly.
My old RAC was around 80-100k, now it should be around 500-600k doing single task Theory's.
My RAM is sure getting a workout though.

Seems to only be 'vbox' Theory's, not the native app. (I haven't tested the vbox on the dev site yet)
I can't really suggest a way to fix this.
Kind of makes the years of crunching only 2 tasks per host feel meaningless lol.
ID: 39503 · Report as offensive     Reply Quote
Profile MAGIC Quantum Mechanic
Avatar

Send message
Joined: 24 Oct 04
Posts: 839
Credit: 36,634,197
RAC: 35,396
Message 39504 - Posted: 7 Aug 2019, 4:50:47 UTC - in response to Message 39503.  

I only saw this on VB version since I don't run any of the natives here or at dev.

lol my Rac here would be just like yours is if I was getting 5000 - 23,000 credits per single core VB task
ID: 39504 · Report as offensive     Reply Quote
Profile MAGIC Quantum Mechanic
Avatar

Send message
Joined: 24 Oct 04
Posts: 839
Credit: 36,634,197
RAC: 35,396
Message 39669 - Posted: 20 Aug 2019, 21:05:56 UTC

https://lhcathome.cern.ch/lhcathome/results.php?hostid=10360630

Single core tasks getting 30+ times the credits as a 2-core of the very same tasks......or thousands of other single core tasks.

Explains the fake Rac's
ID: 39669 · Report as offensive     Reply Quote

Message boards : Theory Application : strange discrepency in credit points for Theory


©2019 CERN