Message boards :
Number crunching :
Astropulse V6 estimates four times too long.
Message board moderation
Author | Message |
---|---|
alan Send message Joined: 18 Feb 00 Posts: 131 Credit: 401,606 RAC: 0 |
I've done 13 Astropulse V6 units now but BOINC still estimates four times too long for new ones (8 days instead of 2 days). This makes setting limits for task fetch a bit tricky. Two astropulse units freeze all further task fetching. I only upgraded to BOINC 7.0.25 from 6.12.34 a few days ago, and have only done one Astropulse unit since then. Is there anything I can do to improve the estimate for Astropulse units? The estimates for ordinary units are pretty accurate. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
I've done 13 Astropulse V6 units now but BOINC still estimates four times too long for new ones (8 days instead of 2 days). This makes setting limits for task fetch a bit tricky. Two astropulse units freeze all further task fetching. Unfortunately, all you can do is wait. The task estimation scheme in use now requires that you complete 10 tasks for any new application before the estimates settle down to realistic values. And those 10 tasks have not only to be validated by your quorum partner, but also mustn't suffer from any of the complications which affect run time (too much radar blanking, or an early exit because too many pulses are found). The good news is that your application details for host 6100462 show that you have nine qualifying tasks already - one more home run, and you'll be out of the woods. |
alan Send message Joined: 18 Feb 00 Posts: 131 Credit: 401,606 RAC: 0 |
Interesting - that page shows the 13 consecutive valid tasks that I have in my log, but only 9 of them are "complete". What's the difference likely to be? All 13 have been validated and awarded credit. Do I have to complete ten tasks under BOINC 7.0.25, or will the ones done using BOINC 6.12.34 count too? Also interesting is that it shows 782 tasks completed for MB but only 388 consecutive valid. I've never had an invalid result, and can only remember a couple of inconclusive ones that were subsequently validated. Would they interrupt a run of consecutive valid results? If there's someplace I can look all this up for myself, point me there and I can stop bothering you. Or just tell me to go away :) Oh, and thanks for the quick reply :) |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Interesting - that page shows the 13 consecutive valid tasks that I have in my log, but only 9 of them are "complete". What's the difference likely to be? All 13 have been validated and awarded credit. The short running MB -9 tasks, and AP tasks with more than 10% Blanking (as well as tasks that finish with 30 pulses and 30 repeating pulses) aren't included in the 'Number of tasks completed' figure, only full length normal tasks. Claggy |
alan Send message Joined: 18 Feb 00 Posts: 131 Credit: 401,606 RAC: 0 |
Thanks for the extra info. Would those terminate with an error or finish and validate as other tasks do? If they don't terminate with an error code how would I recognise them? |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Thanks for the extra info. No they validate normally. With each crime and every kindness we birth our future. |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
Thanks for the extra info. AstroPulse on ATI 5870 GPU. Host 6628794. Even heavily blanked AstroPulseWU's do (reasonable) well on ATI 5870 GPUs. |
ChiTownDale Send message Joined: 18 May 99 Posts: 3 Credit: 284,989 RAC: 0 |
Your experience is different from mine with Astropulse WU's. I'm currently processing two WU's and they are different from each other as night is to day. On one of the WU's it is running maybe 15-20% less hours than the original download's estimate. They both estimated about 65 hours or so. It has clocked 24Hrs 50Min and indicates there are about 33Hrs 50Min of processing left. So it looks like it will take less than 59Hrs to complete the WU. The other one is quite a different story. It has already clocked 67Hrs 40Min of CPU time and shows that there are still 78Hrs 30 Min left to process BUT the amount of time remaining to process the WU is continuing to INCREASE, not decrease just as it has done since the WU was downloaded since it has already used more than the original estimate of 65Hrs to complete the WU. It can't be due to one WU getting more CPU time than the others since I run a dual processor plus a GPU on my display board which processes all of the SETI WU's that can be processed by a GPU processor. So while one has an estimate that may come out using 10-15% less CPU time than estimated, the other one will end up using far in excess of 225% of the original estimate, assuming it actually completes the WU by 4:24 PM on 6/23/12, which is now only 9 days from now. As long as the estimated time required continues to increase instead of decreasing, likelihood that it will actually complete its task gets dimmer and dimmer since at this moment it has already used 2.82 days of CPU time over the past couple of weeks (after all, I do use my computer for more than just processing SETI WU's) and with 3.27 days of CPU time needed to complete the WU and that amount increasing steadily, it is obvious that there is astrong possibility that it nay never complete this task at all. After all, I do also process data for ClimatePredict and those WU's usually require between 190-350 hours of CPU time but they allow me as much as a year to complete one WU so I can SUSPEND processing for months at a time while processing other BOINC projects that have more immediate deadlines to be met. In addition, ClimatePredict grants credit for the portion of a WU that is completed as that portion is reported on an interim bases because they require so much CPU time to complete and even partially completed WU's have some value to their overall projects. I don't know if that is true with Astropulse WU's but if they are going to use over 100 hours of CPU time they really should grant credit for large enough segments of the WU that are completed particularly since the number of weeks granted to complete each WU is such a small fraction of the time ClimatePredict allows for completion of a WU yet apparently for at least some of Astropulse's WU's require nearly as many hours of CPU as ClimatePredict's do. |
Juha Send message Joined: 7 Mar 04 Posts: 388 Credit: 1,857,738 RAC: 0 |
The other one is quite a different story. It has already clocked 67Hrs 40Min of CPU time and shows that there are still 78Hrs 30 Min left to process BUT the amount of time remaining to process the WU is continuing to INCREASE, not decrease just as it has done since the WU was downloaded since it has already used more than the original estimate of 65Hrs to complete the WU. That workunit has probably gotten stuck. The progress meter isn't advancing either, is it? Try restarting BOINC (in BOINC Manager File -> Exit -> Shutdown client (sorry, don't remember the exact wording)) or if that fails, try rebooting. Once you restart BOINC that task should continue to run normally. edit: In 6.12.34 that selection is called "Stop running science applications when exiting the Manager". In your version likely very similar. Somewhat off-topic: Your GPU seems to be producing a few inconclusive results. Your GPU finds 31 signals while your wingmen finds less. I don't know much about GPUs so you'll need to ask advice from someone else. |
alan Send message Joined: 18 Feb 00 Posts: 131 Credit: 401,606 RAC: 0 |
Well, I now have my 10 tasks completed (with 15 consecutive valid tasks; still don't understand that) but I haven't been able to get any more AP units since 31st May, so I don't know if my estimated times have become more sensible. I don't know if this is associated with the wait for the last Astropulse 5 units to clear the system, but there seem to be no Astropulse units available? |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
Well, I now have my 10 tasks completed (with 15 consecutive valid tasks; still don't understand that) but I haven't been able to get any more AP units since 31st May, so I don't know if my estimated times have become more sensible. I don't know if this is associated with the wait for the last Astropulse 5 units to clear the system, but there seem to be no Astropulse units available? At the moment, from the SERVER page : SETI@home server status information is also available in XML. [As of 15 Jun 2012 | 21:20:06 UTC] MB AstroPulse 18se10ag 50.20 GB (12) (done) 23mr10ab 50.20 GB (7) (done) 24ap10ae 50.20 GB (5) (done) 25jn10aa 50.07 GB (3) (done) 25mr10ag 50.20 GB (done) 25se10af 0.00 GB (done) 28jn10ab 50.07 GB (done) 31oc10aa 50.20 GB (done) total available channels on disk:112 0 channels in progress: 1 0 channels done: 26 0 channels ended in error:0 0 total channels to do: 85 0 |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
Got an AP unit which lasted a few seconds, was uploaded and validated. It got me 0.04 credits. Too much RFI, I believe. Tullio |
James Sotherden Send message Joined: 16 May 99 Posts: 10436 Credit: 110,373,059 RAC: 54 |
My P4 is showing AP times of 242 hours to completion. They should take about 37 hours or so running Lunatics. I show 2 completed and validated. Do I have to do 8 more to get the times down again. Im running a 3 day cache and that is the only work unit. I know I have done my 10 completed and validated some months ago when they went to the new version. Im not going to loose sleep over this, but what happend to make my times go over board? [/quote] Old James |
Horacio Send message Joined: 14 Jan 00 Posts: 536 Credit: 75,967,266 RAC: 0 |
My P4 is showing AP times of 242 hours to completion. They should take about 37 hours or so running Lunatics. I show 2 completed and validated. Do I have to do 8 more to get the times down again. Im running a 3 day cache and that is the only work unit. As the APR for APs in that host seems right, probably the previuos completed WU was longer than expected and rised the DCF... |
alan Send message Joined: 18 Feb 00 Posts: 131 Credit: 401,606 RAC: 0 |
You have to laugh. No, really, you do. Tears are not manly. Ten tasks completed, so BOINC should be able to estimate the workunit times a bit better, yes? Finally got another AP workunit. Now, instead of eight days (when it should be two) - - - it estimates ten. Yes, it's got worse, not better. Unless this is some kind of unannounced super-large AP unit, I've got a problem here. Running Astropulse v6 (of course), latest Lunatics optimised version for Windows. |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
My latest AP estimate is 32 days. It's running high priority, of course, so that Test4Theory@home won't start because it needs all of my two cores. Luckily LHC@home has gone to normal priority after reaching 33% in 10 hours, so I won't have 2 high priority runs grabbing all of my 2 cores. Tullio |
alan Send message Joined: 18 Feb 00 Posts: 131 Credit: 401,606 RAC: 0 |
It's now re-estimated the new AP units, presumably based on the times for my recent MB units, and has upped the estimate to 14 days, seven times too large (unless these are some new kind of unit). They are trying to run High Priority, so I'm suspending them while I deal with the wingmen who've returned results and are waiting for me. What's wrong here? I've qualified for better estimates, returning ten units completed, and fifteen consecutive valid tasks, all in less than two days each. In the application details, even Seti@home gives my estimated turnaround time as 7.1 days. Why is BOINC so far up the creek? Is it because I'm running an "anonymous platform"? |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
You have to laugh. No, really, you do. Tears are not manly. The estimate is based on 10 similar blanking tasks, not just 10 tasks. ;) Cheers. |
©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.