Message boards : Number crunching : Credit varies by factor of one hundred for equal CPU time
Author | Message |
---|---|
BKFC Send message Joined: 21 Apr 20 Posts: 34 Credit: 3,160,585 RAC: 0 |
The credit system is not a big deal for me, but I realize that I don't understand the algorithm. Is it discovery based? 1257063659 1126740080 4201228 9 Sep 2020, 18:18:12 UTC 11 Sep 2020, 3:35:51 UTC Completed and validated 29,244.00 28,795.92 368.05 Rosetta v4.20 x86_64-pc-linux-gnu 1257061991 1126738488 4201228 9 Sep 2020, 18:14:00 UTC 11 Sep 2020, 0:08:21 UTC Completed and validated 29,225.16 28,783.33 233.96 Rosetta v4.20 x86_64-pc-linux-gnu 1257061755 1126738301 4201228 9 Sep 2020, 18:14:00 UTC 11 Sep 2020, 2:17:39 UTC Completed and validated 29,258.01 28,797.22 229.21 Rosetta v4.20 x86_64-pc-linux-gnu 1257061811 1126738300 4201228 9 Sep 2020, 18:14:00 UTC 11 Sep 2020, 2:17:39 UTC Completed and validated 29,231.68 28,789.95 316.75 Rosetta v4.20 x86_64-pc-linux-gnu 1257021474 1126702466 4201228 9 Sep 2020, 16:35:42 UTC 10 Sep 2020, 22:33:41 UTC Completed and validated 29,180.66 28,790.75 410.66 Rosetta v4.20 x86_64-pc-linux-gnu 1257021530 1126702477 4201228 9 Sep 2020, 16:35:42 UTC 10 Sep 2020, 22:33:41 UTC Completed and validated 30,181.75 29,392.38 4.03 Rosetta v4.20 x86_64-pc-linux-gnu |
Bryn Mawr Send message Joined: 26 Dec 18 Posts: 393 Credit: 12,114,842 RAC: 4,200 |
The credit system is not a big deal for me, but I realize that I don't understand the algorithm. Is it discovery based? Rosetta is an unusual project in that work units run for a fixed time rather than do a fixed amount of work. The amount of work that your computer does in that time can vary greatly and the credits can reflect this. There can also be a settling in period for new installations or new apps as Boinc works out the work being done. If you look inside the work unit you will see details of the number of decoys completed. For a given starting position the number of decoys is a good measure of work done but it depends on the complexity of the protein being processed and so even this is not a strict measure of the credits you should receive. As you can tell from this ramble I don’t fully understand the credit system, it is an arcane process at the best of times, I tend to put a single WU with an odd credit score down to life’s experience, shrug, and walk away :-) |
BKFC Send message Joined: 21 Apr 20 Posts: 34 Credit: 3,160,585 RAC: 0 |
Thanks for that input. There are still some principles that I don't understand. I have two tasks whose credit differs by 400:4 but have almost identical CPU time - not wall clock time - so I assume neither is sitting idle. The task with 4 credits also had a small number of decoys, but that begs the question as to how credits are awarded, since what's actually donated by volunteers is some number of floating point operations rather than a 'product' (number of decoys?) over which the volunteers have no control. I realize that this is not quite an idle question. I limit CPU resources for this since running the Linux box 24/7 over long periods causes it to be hotter than I would like. But depending upon how I set limits, I could inadvertently set up situations where I get tasks that run very inefficiently (few decoys/GFLOP). Without really knowing the algorithm, it's hard to say. |
Bryn Mawr Send message Joined: 26 Dec 18 Posts: 393 Credit: 12,114,842 RAC: 4,200 |
OK, I’ve just had a look at your setup and the first thing I would say is, unless you have a really unreliable ISP, cut down your work cache. A setting of 0.1+0.2 day’s work is far more manageable. Looking at the WU that was awarded 4.03 credits I think I can see why. It failed, twice, and called boinc-finish three times reporting 46, 1 and 1 decoy. Whenever this happens the credit scoring system appears to ignore all but the last finish command so you were credited for 1 erroneous WU and the 46 good WUs were lost. |
Brian Nixon Send message Joined: 12 Apr 20 Posts: 293 Credit: 8,432,366 RAC: 0 |
The one that only got 4 credits looks like a bug which crops up occasionally: somehow the task runs more than once, and seemingly only the last run (which was very short) gets any credit, even though it was the first run that did almost all the work. |
Message boards :
Number crunching :
Credit varies by factor of one hundred for equal CPU time
©2024 University of Washington
https://www.bakerlab.org