Message boards :
Number crunching :
'to completion' times grossly overestimated.
Message board moderation
Author | Message |
---|---|
Kaylie Send message Joined: 26 Jul 08 Posts: 39 Credit: 333,106 RAC: 0 |
estimated completion times for WU's are being estimated at 35 hours when the tasks actually finish in seven. This seems to limit the amount of work BOINC will accept. Any ideas how to fix this? |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
estimated completion times for WU's are being estimated at 35 hours when the tasks actually finish in seven. This seems to limit the amount of work BOINC will accept. Jut wait. Boinc learns from experience so the estimates will improve quite quickly. F. |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
estimated completion times for WU's are being estimated at 35 hours when the tasks actually finish in seven. This seems to limit the amount of work BOINC will accept. It'll fix itself after time, you might just have to give it a few days. It's called the Task Duration Correction Factor, basically it's a calculation of the difference between how long BOINC thinks it will take to complte a task and how long it actually does take. To check exactly what your Task Duration Correction Factor is, go to Your Account > Your Computers > Click "Details" on one of them > And scroll down to the bottom and above "Location" and "Merge This Computer" is your TDCF. I think the optimum is 1.00? My two computers are showing 0.55 and 0.11 respectively. - Luke. - Luke. |
Kaylie Send message Joined: 26 Jul 08 Posts: 39 Credit: 333,106 RAC: 0 |
for quite some time the estimates have been fairly acurate. now they drop slightly when tasks finish but go right back up again a short time later. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
for quite some time the estimates have been fairly acurate. now they drop slightly when tasks finish but go right back up again a short time later. Before you decide this is a huge problem, you need to look at how BOINC uses the time estimates, and how inaccurate estimates could cause problems. Let's say you want a 7 day cache, and each work unit takes about ten hours (just for easy math). 168 hours in a week, BOINC rounds up, so 17 work units in your queue. If BOINC over estimates, it will carry less work. If BOINC estimates 100 hours per work unit, you'll have two in your queue, and while the queue isn't very full, the risk of missing deadlines is small, because that 200 hours of queued work is really 20. On the other hand, let's say that BOINC under estimated, and thought the work would take 1 hour when it takes ten. Now your queue is 168 work units, and instead of being one week, it is ten weeks. Now you're in trouble. So the estimates are intentionally high. As others have been pointing out, there is a correction factor that BOINC uses to bring estimates into line. It is intentionally intended to adjust downward very slowly, and adjust upward quickly, because under-estimates are really bad, while over-estimates are mostly cosmetic. |
perryjay Send message Joined: 20 Aug 02 Posts: 3377 Credit: 20,676,751 RAC: 0 |
Since your computers are hidden it's kinda hard to tell but if you are running a CUDA card that could be messing with your DCF. We have been getting a lot of shorties lately and if they have been running on your GPU they will drag the DCF way down. Then when you turn in a couple of midrange WUs with your CPU DCF will go crazy trying to figure out what happened. PROUD MEMBER OF Team Starfire World BOINC |
RB Send message Joined: 7 Mar 00 Posts: 103 Credit: 1,084,436 RAC: 0 |
I think the optimum is 1.00? My TDCF is 0.168309 The WU's complete in almost the exact time that is shown in the estimate. (within a minute or so, if not exact) Shouldn't my number be closer to 1.00 then, for being so close? Just wondering... |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I think the optimum is 1.00? 1.00 is the default value. The more work your machine can do the lower boinc will make this value. 0.15488 is the value for my 3.0GHz E8400 C2D 0.684242 is the value for my triple Pentium II 400MHz server SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
hiamps Send message Joined: 23 May 99 Posts: 4292 Credit: 72,971,319 RAC: 0 |
estimated completion times for WU's are being estimated at 35 hours when the tasks actually finish in seven. This seems to limit the amount of work BOINC will accept. On my i7 it is Task duration correction factor 0.08962 and it has lots of units. Official Abuser of Boinc Buttons... And no good credit hound! |
Kaylie Send message Joined: 26 Jul 08 Posts: 39 Credit: 333,106 RAC: 0 |
Still over 30 hours for 7 hour tasks on a quad core with two GPU's (there have been some short CUDA's) and the DCF for that computer is an outragous 2.9 I understand the need to insure meeting the deadlines. |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
Still over 30 hours for 7 hour tasks on a quad core with two GPU's (there have been some short CUDA's) and Yep. Give it time to get lower. You might need a week or more for it to fully stabilize, as it is slow to come down and fast to rise up. - Luke. - Luke. |
Ianab Send message Joined: 11 Jun 08 Posts: 732 Credit: 20,635,586 RAC: 5 |
Another thing I have noticed is that if a PC is put into "Suspend" mode it seems to mess with the correction factor. Not sure of the details, but one of my P4s got suspended for a couple of days and the DCF went from it's normal 0.25 to 0.8 So although it's set to 3 days work, it was only requesting about 1 day. As a work around, increase the work unit cache temporarily, that will fool it into downloading some more work units and it will correct itself as work units are returned with the correct time calculations. Ian |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Still over 30 hours for 7 hour tasks on a quad core with two GPU's (there have been some short CUDA's) and Not sure about the outrageous part. You can go and find the <duration_correction_factor> tag in client_state.xml (stop BOINC before editing) and change it if you wish (0.5 is probably a safe number), or as others have suggested, you can wait. Each time you finish a WU, it'll go down a bit. |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
Still over 30 hours for 7 hour tasks on a quad core with two GPU's (there have been some short CUDA's) and Outrageous was the gross underestimate of time that some projects have had. I had a DCF on one projct of 1E30 (Not a typo that is a 1 with 30 zeros after it). Problems like this got the DCF capped at 100, and any DCF over 90 would automatically put ALL tasks for that project into Earliest Deadline First (at this point, since BOINC has capped the DCF, it has NO idea what is happening with time for that project). BOINC WIKI |
Kaylie Send message Joined: 26 Jul 08 Posts: 39 Credit: 333,106 RAC: 0 |
It's happening again, just when I want to load up with tasks for the outage. I editied the XML file and set TDCF to .5 and the task estimates droped to 8 hrs, when they take 5. That is entirely acceptable but as soon as one 5 hour task finished the estimates on all tasks junped to 27 hours. This sure is annoying... Running BOINC 6.10.17 on a quad core with 2- GPU's. |
Pappa Send message Joined: 9 Jan 00 Posts: 2562 Credit: 12,301,681 RAC: 0 |
It's happening again, just when I want to load up with tasks for the outage. I editied the XML file and set TDCF to .5 and the task estimates droped to 8 hrs, when they take 5. That is entirely acceptable but as soon as one 5 hour task finished the estimates on all tasks junped to 27 hours. This sure is annoying... It sounds like you got a couple of VLARs which drive DCF UP almost instantly. While other more normal WU's return it more slowly... You have two choices: Wait it out. or The only real way to correct that before the outage is do a Project Detach and then Attach... It will clean the information in that is in the client_state.xml file. Regards Please consider a Donation to the Seti Project. |
S@NL - eFMer - efmer.com/boinc Send message Joined: 7 Jun 99 Posts: 512 Credit: 148,746,305 RAC: 0 |
It's happening again, just when I want to load up with tasks for the outage. I editied the XML file and set TDCF to .5 and the task estimates droped to 8 hrs, when they take 5. That is entirely acceptable but as soon as one 5 hour task finished the estimates on all tasks junped to 27 hours. This sure is annoying... BOINC can't handle CPU and GPU right, at least not the latest stable release. Setting the FLOPS statement solved this problem on my computers. Is goes wrong when the GPU is a lot faster than the CPU. Make the FLOPS value larger when the GPU tasks take shorter than predicted. (GTX 295) one half. <app_version> <app_name>setiathome_enhanced</app_name> <version_num>608</version_num> <flops>22000000000</flops> <plan_class>cuda</plan_class> <avg_ncpus>0.040000</avg_ncpus> <max_ncpus>0.040000</max_ncpus> - <coproc> <type>CUDA</type> <count>1</count> </coproc> - <file_ref> <file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048.exe</file_name> <main_program /> </file_ref> - <file_ref> <file_name>cudart.dll</file_name> </file_ref> - <file_ref> <file_name>cufft.dll</file_name> </file_ref> - <file_ref> <file_name>libfftw3f-3-1-1a_upx.dll</file_name> </file_ref> </app_version> TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
It's happening again, just when I want to load up with tasks for the outage. I editied the XML file and set TDCF to .5 and the task estimates droped to 8 hrs, when they take 5. That is entirely acceptable but as soon as one 5 hour task finished the estimates on all tasks junped to 27 hours. This sure is annoying... For CPU work, BOINC keeps a <duration_correction_factor> which adjusts for the difference between the benchmark prediction, and reality. BOINC needs a DCF for the GPU as well, but the real problem is that there is no suitable GPU benchmark. ... and as Fred points out, GPUs vary wildly. In the grand scheme of things, this is mostly cosmetic. As long as BOINC doesn't grossly overestimate performance, work will be completed on-time. |
S@NL - eFMer - efmer.com/boinc Send message Joined: 7 Jun 99 Posts: 512 Credit: 148,746,305 RAC: 0 |
It's happening again, just when I want to load up with tasks for the outage. I editied the XML file and set TDCF to .5 and the task estimates droped to 8 hrs, when they take 5. That is entirely acceptable but as soon as one 5 hour task finished the estimates on all tasks junped to 27 hours. This sure is annoying... I know the Lunatics guys are working on this. TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking. |
Kaylie Send message Joined: 26 Jul 08 Posts: 39 Credit: 333,106 RAC: 0 |
Think I see now what is happening. My GPU's run slower than estimated and, scince there is only one TDCF for both CPU and GPU, the CPU task length estimates suffer. This make sense? |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.