Message boards :
Number crunching :
Qsqrt421_DS3x8 15+ hour running time?
Message board moderation
Author | Message |
---|---|
Send message Joined: 24 Aug 15 Posts: 3 Credit: 50,248 RAC: 0 |
Just finished the first one second one is reporting soon http://numberfields.asu.edu/NumberFields/result.php?resultid=15861831 |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
Just finished the first one second one is reporting soon http://numberfields.asu.edu/NumberFields/result.php?resultid=15861831 Ignoring the handfull of really long WUs, 15+ hours is not uncommon, but most should be less than that. |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
I don't know if this can be called "a handful". I've got an entire bunch of really long running ones, even on really fast machines like an i7-3820 at 4,4Ghz. These workunits will miss the deadline soon, you should think about extending the deadlines in general. I let these workunits run but I will be very disappointed when I can't get credit for it. http://numberfields.asu.edu/NumberFields/result.php?resultid=16168339 http://numberfields.asu.edu/NumberFields/result.php?resultid=16168357 http://numberfields.asu.edu/NumberFields/result.php?resultid=16168378 |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
I don't know if this can be called "a handful". The ones you listed should finish within a few days. The worst I have seen on my computer is 80 hours; it's a small sample of the WU population, but I don't expect it to be much worse. The deadline and grace period is already set pretty high, so I would rather not adjust that anymore. If you miss the deadline and someone finishes the WU before you, I can manually grant you the credit, so don't worry about that. The "handful" I was referring to were the ones that were taking between 2 and 4 weeks (sometimes longer depending on the host). These were extensively discussed in other threads. It turns out you do have a couple of those: http://numberfields.asu.edu/NumberFields/result.php?resultid=16374076 http://numberfields.asu.edu/NumberFields/result.php?resultid=16197688 I have been monitoring these and there are only 8 left. When someone finishes one, I alert the other user with the same WU and give them manual credit for their trouble. |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
Okay, that's good enough for me. One of the i7-3820's WUs I listed is now running for a week and that's a very fast machine on NumberFields. someone finishes the WU before you I don't think so... ;) |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
The first two of my long running WUs I listed yesterday are down. Unfortunately, it found nothing. # Inspected 2731542253 polynomials. But I've reconized another problem regarding Boinc credits with these WUs. http://numberfields.asu.edu/NumberFields/result.php?resultid=16168378 ran for almost a week and scored 5,745.30 credits http://numberfields.asu.edu/NumberFields/result.php?resultid=16168357 ran for 4 days and 7 hours, scoring 5,745.30 credits too Shouldn't there be a difference between these two because of the 2 1/2 day longer running time? I assume that these WUs reach an upper limit in the credit calculation process. |
Send message Joined: 11 Apr 15 Posts: 4 Credit: 4,721,492 RAC: 1,147 |
I also have seen this as of today. One WU consumed 373,000 CPU seconds and another 184,000 seconds, but both were granted 4,069 credits. http://numberfields.asu.edu/NumberFields/result.php?resultid=16170600 http://numberfields.asu.edu/NumberFields/result.php?resultid=16169189 As long as I'm here, let me compliment the project. This project has a very good level of communication and friendly responsiveness from the administrators, and I especially appreciate the batch status page, as I enjoy keeping track of the progress. Some projects are black holes that consume all the CPU you give them with no sense of progress. So, congratulations to this crew for providing that. |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
The first two of my long running WUs I listed yesterday are down. Unfortunately, it found nothing. Finding no fields is ok - some regions of the search space contain no fields. This is the norm for the regular decics app. The bounded app on the other hand will usually find fields; this is especially true for the current subfield under test. Regarding the credits, I think the problem might be the credit cap. This was implemented a while back to thwart "cheaters" who find ways to artificially increase their run times. Let me see what I can do about the cap. |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
The first two of my long running WUs I listed yesterday are down. Unfortunately, it found nothing. So the problem was the cap. I temporarily increased the cap until run times return to normal. I also updated your credits to what they should have been. Let me know if there are any others. |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
I also have seen this as of today. One WU consumed 373,000 CPU seconds and another 184,000 seconds, but both were granted 4,069 credits. I also updated your credits to what they should have been (the 2nd task was essentially at the cap, so no change there). And thanks for the compliment! |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
Some projects are black holes that consume all the CPU you give them with no sense of progress. That's true for most of the projects. Aside a good progress indicator (which virtually stands still at large WU batches) I like to know if I found something important. That's why I'm inspecting WU output files from extraordinary WUs or when I reconize an unusual file size. On Primegrid you have a list of all primes you found. I don't know how many fields or "important polys" are found during the process or within a normal day, but if they are very rare such a list (or a counter when not that rare) will be nice here too. Don't care that much about cheaters. If they need more credits than others for compensating their bad live, they won't stop even if there is an upper limit. This limit gets worked out then maxed out without beeing reconized by the script. My approach to counter this would be calculating how much polys where tested over time and comparing this against processor capabilities. Even I won't grant credits based on benchmark/time relations, but for tested polys. This will give the same amount of credits for the same amount of work regardless of which processor has finished the WU or how good or bad the benchmark results have been. |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
The tasks you listed as "heavy" on my i7 machine have both started. While #16197688 looks not this much disturbing for now (47.463% after 92:12:00), #16374076 could provide some fun. It's hanging around at 0.038% after 22:36:00, resulting in an estimated finishing time in a couple of years. Well, let's see. My plan is to let them run, hoping the max_time and max_ops values are set high enough to not get the WUs killed by the Boinc Manager. Fortunately the machine is stable and really fast at NumberFiels and both tasks are on the same machine. I should be able to complete these WUs. |
Send message Joined: 28 Oct 11 Posts: 179 Credit: 223,858,438 RAC: 112,857 |
I have been monitoring these and there are only 8 left. When someone finishes one, I alert the other user with the same WU and give them manual credit for their trouble. Looks like you might be down to 7 - my copy of WU 12404860 is back, after 46 days. Thanks for your email the other day: I did abort the spare copy I was running as you suggested. But the email address you used doesn't seem to be accepting replies - it bounced with "Mail server for "mimas.la.asu.edu" unreachable for too long". So now I'm down to one - waiting to see which one of us on WU 12420121 finishes first. |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
In this case I want to abort #16197688. Or do you want to double-check it? Just doing such a WU for fun doesn't seem efficient to me. |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
In this case I want to abort #16197688. Or do you want to double-check it? Yes, go ahead and abort that one. I will grant you the canonical credit for your trouble. Thanks! |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
Due to the war on SPAM, the ASU IT department uses some really strict mail policies. In fact, I don't think mimas even has a mail server running. It was a pain just to get private messaging working - I can't remember exactly how I did that but as far as I know it works (at least I can send/receive pms). |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
Okay thank you. But let me finish the other one! |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
Some projects are black holes that consume all the CPU you give them with no sense of progress. That's an interesting idea for computing credits. I think the main obstacle to doing that is that it doesn't play well with the built in BOINC crediting mechanisms, so I would have to modify a bunch of code to make it work. Then of course the cheaters will find ways to modify the output scripts to make it look like they tested more polynomials than they really did. Finally, the number of tested polynomials is not necessarily a good measure of how many cpu cycles the host donated; some may find that unfair. |
Send message Joined: 8 Jul 11 Posts: 1323 Credit: 410,949,970 RAC: 246,377 |
Okay thank you. But let me finish the other one! Please do! |
Send message Joined: 23 Feb 13 Posts: 29 Credit: 21,480,710 RAC: 0 |
sorry... but credit cap is still too low for some tasks... http://numberfields.asu.edu/NumberFields/result.php?resultid=16168410 http://numberfields.asu.edu/NumberFields/result.php?resultid=16168407 http://numberfields.asu.edu/NumberFields/result.php?resultid=16168395 |