Message boards :
Number crunching :
Long WU's
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 · Next
Author | Message |
---|---|
Send message Joined: 27 Oct 07 Posts: 186 Credit: 3,297,640 RAC: 0 ![]() |
I'm wondering what limit Igor might have set when he implemented credit based on runtime. Maybe the maximum runtime for which credit will be granted was reasonable for the previous tasks, but needs a nudge upwards to suit this new experiment. FWIW, I'm seeing the normal credit per hour from all jobs, but I haven't completed/validated any really long ones yet. I'll keep an eye on them as they report in. |
![]() Send message Joined: 4 May 07 Posts: 250 Credit: 826,541 RAC: 0 |
I'm wondering what limit Igor might have set when he implemented credit based on runtime. Here is one of my long jobs that has completed and validated. http://lhcathomeclassic.cern.ch/sixtrack/workunit.php?wuid=2518553 And another one... http://lhcathomeclassic.cern.ch/sixtrack/workunit.php?wuid=2495733 I'm computer 9955823 |
![]() Send message Joined: 2 Sep 04 Posts: 309 Credit: 715,258 RAC: 0 |
|
Send message Joined: 12 Jul 11 Posts: 857 Credit: 1,619,050 RAC: 0 |
Thanks for the useful feedback. We shall look whenever we can. Eric. |
![]() ![]() Send message Joined: 30 May 08 Posts: 93 Credit: 5,160,246 RAC: 0 ![]() ![]() |
FWIW, I'm seeing the normal credit per hour from all jobs... Unfortunately, I'm not having the same experience. All of my hosts have crunched some of these monster WU's and the long ones are receiving about 25%-33% of the credits/hour of the shorter tasks. For example from my i5: 5799577 350 sec 21.6 cr/hr 5502249 12K sec 27.5 cr/hr 5464672 103K sec 7.4 cr/hr |
Send message Joined: 27 Oct 07 Posts: 186 Credit: 3,297,640 RAC: 0 ![]() |
FWIW, I'm seeing the normal credit per hour from all jobs... Yes, that's what I meant and expected. I've been away from the project for a week, crunching SIMAP, so although I've been allocated long WUs, the only ones I've been able to report have been the ones which finish early. I've got several crunching at about 1% an hour, but so far they've only reach 20% after 20 hours. It'll be another three or four days before I can repeat your observation. |
Send message Joined: 27 Oct 07 Posts: 186 Credit: 3,297,640 RAC: 0 ![]() |
Another looming problem: I'm starting to get long WU resends, with the accelerated deadline of less than 3.5 days. Fortunately, my first two have been short-runners, but heaven forfend I get a full 100-hour job with a half deadline.... |
Send message Joined: 27 Sep 08 Posts: 817 Credit: 677,569,806 RAC: 219,582 ![]() ![]() ![]() |
re: ritterm I'm seeing about the same, the jobs that are >24hrs see a drop in cr/hr Credit CPU T cr/hr 1.85 0.09 18.5 138.19 7.55 18.3 182.57 15.3 11.9 182.57 19.0 9.6 182.57 24.0 7.6 316.69 36.5 8.7 To be honest it's probally something weired with credit new as it doesn't know about these long WU's: Credit CPU T cr/hr 316.69 36.5 8.7 182.57 38.4 4.7 182.57 27.9 6.5 268.58 56.9 4.7 221.6 67.4 3.3 You can see in rough assending date that the cr/hr is increasing |
Send message Joined: 12 Jul 11 Posts: 857 Credit: 1,619,050 RAC: 0 |
Right; this is not satisfactory. I'll fix that soonest (I hope). Eric. |
![]() Send message Joined: 3 Mar 11 Posts: 3 Credit: 560,746 RAC: 0 |
it's great to know what is the problem (finding out there's NO problem), but seriously, no credits for such WUs will earn you a serious fall in project participation. I'm a somewhat devout LHC cruncher, while not great, and seeing this: http://lhcathomeclassic.cern.ch/sixtrack/workunit.php?wuid=2481017 with 260k seconds earn me 277cr is greatly discouraging. Still it's good to know this is being addressed. By the way, you can shift to fixed credits for 10m turns runs once they pass validation, and some fixed credits for failed runs too, depending on when they have failed, taking values from the task result (I expect it sends the number of turns when it discovers particles to hit the wall). |
![]() Send message Joined: 12 Dec 05 Posts: 31 Credit: 9,709,398 RAC: 0 ![]() ![]() |
if you're in it for the points (moreso than you are for the science), then you're definitely participating in the wrong project LOL... even if the points system got fixed for these long running tasks, LHC@H is notorious for being stingy w/ points...even on the rare occasion in which the project server dishes out a long, steady flow of new work. all i care about are scientific results, and possibly some recognition if my host ends up being one of the few out there to help make a new discovery, evolution, or revolution in high-energy particle physics. but for all the tasks i crunch that don't end up being something special, i could care less about the points they earn. keep up the good work Eric. Eric ![]() |
![]() Send message Joined: 17 Sep 04 Posts: 104 Credit: 32,486,555 RAC: 6,789 ![]() ![]() ![]() |
if you're in it for the points (moreso than you are for the science), then you're definitely participating in the wrong project LOL.. Everyone is into the project for different reasons, so to maintain maximum participation, a project should do its best to accommodate all of them to the best of its ability, imho. Regards, Bob P. |
![]() Send message Joined: 4 May 07 Posts: 250 Credit: 826,541 RAC: 0 |
if you're in it for the points (moreso than you are for the science), then you're definitely participating in the wrong project LOL... Here! Here! If there is a question about where LHC@Home development time needs to be spent then in IMHO fixing credit calculation shouldn't be high on the priority list. I work Einstein@Home, Seti@Home, Test4Theory@Home (LHC@Gome V2.0), and LHC@Home (V1.0). Because of the high number of Tasks being pushed by LHC@Home I've recently adjusted my Resource share to give priority to LHC@Home. However, LHC@Home hasn't been a large source of work in the past. The sad thing is that BOINC 7.0.25+ has kind of made a mess of work allocation when the Task load between projects isn't well balanced. SETI@Home can quite easily push enough work to swamp Projects with only occasional work. This is even worse because BOINC doesn't understand the difference between CPU and GPU work so a large number of GPU tasks results in the CPU being idle at times. I've ended up using the "arm strong" method of keeping my work queues filled and balanced. |
![]() Send message Joined: 3 Mar 11 Posts: 3 Credit: 560,746 RAC: 0 |
Actually I'm in for another reason than points, I like the LHC idea itself, but points do play a motivational role in participating longer. I believe it's similar with other people interesting in distributed computing. After all, many things today include e-rulers to measure against other people, and if it works, why neglecting it? But with current desolation of cs/hour, even with constant flow of tasks, I've stepped down a little from the project. (To be honest, I do play points-wise, in POEM@home, performing tenfold vs Einstein on my simple GPU) Also, there's a "Formula BOINC" thing that doesn't care about points across projects, and we're in it as a team. |
Send message Joined: 27 Oct 07 Posts: 186 Credit: 3,297,640 RAC: 0 ![]() |
I've got several crunching at about 1% an hour, but so far they've only reached 20% after 20 hours. It'll be another three or four days before I can repeat your observation. My long ones are all back now, though several are still waiting for wingmates to validate. The clearest results have come from host 9845031 (don't worry about the download errors, there was an ISP glitch at my end). Specifically, compare WU 2528002 WU 2527992 In WU 2528002, my...031 host (listed second) took 227 Ksec, but my wingmate was even slower - he will have been affected even harder than me if we assume the credit is capped after a fixed runtime. In WU 2527992, I was paired with a faster host, so my machine will have been more affected by capping, and will have set the credit benchmark for the WU. Both of us got more credit for this WU than I and my partner did for WU 2528002. If I'm setting the rate for WU 2527992, we can work out where the cap bites. This host has been earning a very steady and repeatable 19.61 credits per (runtime) hour since 'credit from runtime' was introduced. 240.55 credits takes exactly 12 hours 16 minutes to earn at 19.61 credits per hour: I think we'll find that the 'credit from runtime' cap has been set at 44,160 seconds. Or have I missed something? |
![]() ![]() Send message Joined: 13 Jul 05 Posts: 143 Credit: 263,300 RAC: 0 |
Just picked up one of the "long tasks", so I look for about an 80 hr run ... we shall see If I've lived this long, I've gotta be that old |
Send message Joined: 27 Jan 08 Posts: 1 Credit: 1,288,004 RAC: 0 |
WU 2476714 157,132.40s CPU @3GHz 188.03 credit ~100cr/day :( wingman: 199,539.50s It's still switching between SSE2 and PNI app on the same cpu |
Send message Joined: 20 Sep 05 Posts: 31 Credit: 1,221,734 RAC: 525 ![]() ![]() |
83 hours gone, still 103 hours to go and already past the expected return date: http://lhcathomeclassic.cern.ch/sixtrack/workunit.php?wuid=2504999 ![]() |
Send message Joined: 4 Sep 05 Posts: 15 Credit: 816,255 RAC: 0 |
I have one that has been running at high priority for 33+ plus hours and constantly shows about 60 hours remaining! It's running on a dual core AMD and the other core (and the GPU) is doing other work OK ... so, I guess I'll just let it keep running until either a) it completes, b) errors out, or c) exceeds its due date of 8/14. ![]() |
Send message Joined: 9 Oct 10 Posts: 77 Credit: 3,671,357 RAC: 0 |
I've just returned one close to the deadline on my i7 ... it almost missed it :( |
©2024 CERN