Questions and Answers :
Unix/Linux :
How is BOINC client scheduler supposed to work?
Message board moderation
Author | Message |
---|---|
Send message Joined: 5 Aug 04 Posts: 1120 Credit: 17,193,804 RAC: 2,852 |
I have three BOINC projects: setiathome, proteinfolding, and climateprediction, set to get 10%, 10% and 80% of the spare time, respectively. My machine has two hyperthreaded Xeon processors, so I say to run at most 4 applications at a time on multiprocessor systems. When running BOINC clients upt to 4.43, the client had lots of problems, but pretty much seemed to be trying to do this. But the new BOINC boinc_5.2.4_i686-pc-linux-gnu client seems to be trying to do those whose due date is nearest. The trouble with this is I keep getting new work units for the setiathome and proteinfolding stuff that have 2-week and 1-week deadlines, so it is not running the climateprediction applications anymore because their deadline is so far out. But not that far: one has 6768 hours to go; deadline January 24: 40.25 weeks one has 7539 hours to go; deadline January 25: 44.75 weeks one has 8752 hours to go; deadline February 14: 52 weeks Now those times to go are by boincmgr, and they seem to be too high. But if those applications do not get scheduled significantly often and for a significant time, they will all be late. |
Send message Joined: 5 Aug 04 Posts: 1283 Credit: 15,824,334 RAC: 0 |
It\'s all explained in the work scheduler entry in the BOINC Wiki. It sounds like your other projects built up a large short term debt while you were running 4.43. I suppose you could always disable work fetch for them to force CPDN to be scheduled. And if you\'re feeling brave you could always stop BOINC and edit client_state.xml to reset the long and short term debts for all projects to start them off on a level playing field again. And yes, BOINC does seriously overestimate the time for all CPDN applications (by up to 50% from what I\'ve seen). It should get better over time as the duration correction factor for the project reduces, but I suspect it\'s going to take at least one model completion before that happens. "The ultimate test of a moral society is the kind of world that it leaves to its children." - Dietrich Bonhoeffer |
Send message Joined: 3 Sep 04 Posts: 268 Credit: 256,045 RAC: 0 |
The problem of the CPDN estimated completion time is worse in Linux than Windows. My Spinup Wu is estimated at 10699 hours instead of 3550 One of my Sulphur Wu is estimated at 5172 hours instead of 1500. So it\'s more than 50% for Linux: I would say 350% for me Arnaud |
Send message Joined: 7 Aug 04 Posts: 2183 Credit: 64,822,615 RAC: 5,275 |
As Thyme states, 5.2.2 ridiculously overestimates time to completion, and it seems to be even worse in Linux than in Windows for whatever reason. Your computers will take nowhere near that long on the CPDN workunits. |
©2024 cpdn.org