Message boards : Number crunching : Rosetta Mini v3.78 - Declining credit since October 24th
Author | Message |
---|---|
jjch Send message Joined: 10 Nov 13 Posts: 14 Credit: 440,529,768 RAC: 21,374 |
I have noticed a significant RAC decline since October 24th running the Rosetta Mini v3.78 application. Refer to my Rosetta Stats here: https://boincstats.com/en/stats/14/user/detail/486414/charts As far as I know everything is working properly and it seems unusual that this would have such a steady drop unless there is a problem somewhere or Rosetta changed something with the 3.78 app. I did see there is a 4.0 app however I have not been receiving tasks for that version yet. |
jjch Send message Joined: 10 Nov 13 Posts: 14 Credit: 440,529,768 RAC: 21,374 |
I am seeing a few 4.0 tasks now but only a very small amount. The RAC is continuing to drop. If this is the new credit model than that's fine. I will just have wait to see when it levels off. If there is something broken I would like to fix it. |
shanen Send message Joined: 16 Apr 14 Posts: 195 Credit: 12,662,308 RAC: 0 |
Not sure if your question might be related to what I noticed about a week ago and commented on (in this community) yesterday. I'm not even sure what your "RAC" stands for, though I guess it has something to do with reported or received credit as a metric of work done. However, I have definitely noticed some strange results: Sometimes a task will run much longer than others do. Most of them run a bit over 8 hours these days (even the rb tasks), but sometimes a task will run up to 11 hours or longer. When I notice such a task, it is almost always in a weird slow state. The remaining time will be something like 10:30 and it will not change. The progress will be something like 98.844% and every so often it may increment by 0.001%, but only a few times per minute. Usually the remaining time decreases as the completion percentage increases, but not in these cases. It would seem that these tasks require extra effort resulting in longer computing times. However the resulting credit is NOT increased to reflect the extra work, but apparently decreased. Seems to feel like some sort of penalty for slowness rather than any reward for persistence and the extra effort. If there are many of those tasks (though right now I don't see any on this computer), then that would obviously reduce the credit earned by the affected machines. Here is an example I noticed yesterday. Most tasks receive a credit around 200. On the current list (for this machine) I see values from 188 to 269 and this one outlier at 119. That is the task that ran for over 12 hours. https://boinc.bakerlab.org/result.php?resultid=955423025 #1 Freedom = (Meaningful - Constrained) Choice{5} != (Beer^3 | Speech) |
shanen Send message Joined: 16 Apr 14 Posts: 195 Credit: 12,662,308 RAC: 0 |
One more addendum on the math. If three normal tasks complete in 24 hours and earn 200 "Work done" points each, that's around 600 points. In contrast, two of the slow tasks could run in 24 hours and only earn 100 points each, for a total of 200. From 600 to 200 looks like a slowing down of 2/3 as measured by that metric. Of course that effect would be limited and I've never seen that many of the slow tasks at one time, but they certainly could be lowering the average when they get mixed in there. #1 Freedom = (Meaningful - Constrained) Choice{5} != (Beer^3 | Speech) |
MindCrime Send message Joined: 28 Feb 14 Posts: 2 Credit: 3,013,860 RAC: 0 |
Not sure if your question might be related to what I noticed about a week ago and commented on (in this community) yesterday. I'm not even sure what your "RAC" stands for, though I guess it has something to do with reported or received credit as a metric of work done. However, I have definitely noticed some strange results: Recent Average Credit, not sure if the boinc server settings for this can be adjusted, it is usually your daily average over the past month. |
Message boards :
Number crunching :
Rosetta Mini v3.78 - Declining credit since October 24th
©2024 University of Washington
https://www.bakerlab.org