Questions and Answers :
Windows :
huge WU
Message board moderation
Author | Message |
---|---|
Send message Joined: 25 Aug 04 Posts: 3 Credit: 41,273 RAC: 0 |
I\'ve just joined this project, and the first unit I\'ve got is HUGE!!! Estimated time to finish it is about 650 hours... It\'s almost a month. Is it ok? SETI sends smaller units... ps. it\'s working on Athlon 3200+ |
Send message Joined: 5 Aug 04 Posts: 85 Credit: 2,924,043 RAC: 0 |
Welcome on board, Navigator! Yes, all CPDN WU takes 3-6 weeks to complete depending on your computer. For your Athlon 3200+, you may expect 3-4 weeks if your run your computer 24/7. Hopefully, your will receive credits at the completion of every 10802 time steps (= 1 trickle) (or appr. at every 8-10 hours of processing). |
Send message Joined: 5 Aug 04 Posts: 1283 Credit: 15,824,334 RAC: 0 |
That's a standard CPDN work unit. The standard work unit has 3*15 year phases. The model works with a standard 30 day month, with 48*30 minute timesteps per day. Processing time for each timestep varies depending on the processor and memory speeds. <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/team_display.php?teamid=3"><img src="http://www.teampicard.net/templates/fisubice/images/phpbb2_logo.jpg"></a> |
Send message Joined: 7 Aug 04 Posts: 2183 Credit: 64,822,615 RAC: 5,275 |
> That's a standard CPDN work unit. > > The standard work unit has 3*25 year phases. The model works with a standard > 30 day month, with 48*30 minute timesteps per day. Processing time for each > timestep varies depending on the processor and memory speeds. > > Must be a typo. Should be 3*15 year phases. Currently 45 model years total. |
Send message Joined: 5 Aug 04 Posts: 1283 Credit: 15,824,334 RAC: 0 |
> Must be a typo. Should be 3*15 year phases. Currently 45 model years total. Just checking how awake everybody is ;) <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/team_display.php?teamid=3"><img src="http://www.teampicard.net/templates/fisubice/images/phpbb2_logo.jpg"></a> |
Send message Joined: 5 Aug 04 Posts: 172 Credit: 4,023,611 RAC: 0 |
I know that the CPDN WUs are big, but I thought that BOINC would do a slightly better job of disallowing slow computers from this project. As part of testing things, I have left a couple of slow machines attached to the project. This afternoon they both got a CPDN WU. The slower of the 2 should be finished in July ----- of 2006 assuming that there are no other projects that it shares with. It has 15531 hours to go, or 92 weeks. jm7 |
Send message Joined: 17 Aug 04 Posts: 753 Credit: 9,804,700 RAC: 0 |
jm7 wrote: > I know that the CPDN WUs are big, but I thought that BOINC would do a slightly > better job of disallowing slow computers from this project. It's a moot point how far CPDN should be trying to prevent slower machines from joining. The main reasons for doing so are: 1) slow machines are unlikely to be capable of completing a model, and are therefore wasting their own and server resources (and board members' time answering queries when they crash :-)) 2) owners of very slow machines are unlikely to have the patience to finish the model so should be saved from themselves. On the first, experience under classic CPDN was that anything less than an 800 MHz P3 was likely to crash the model. We simply don't yet know whether the BOINC version will be more tolerant, but the indications are that it may be. On the second, there is a strong feeling in some quarters that if people are sufficiently motivated to take part that has an educational value in itself. Once they are hooked, there is every chance that they will upgrade, and that has been the experience for quite a number running the project so far. We should prepare for queries on how to transfer models to a new machine! |
Send message Joined: 7 Aug 04 Posts: 187 Credit: 44,163 RAC: 0 |
> It's a moot point how far CPDN should be trying to prevent slower machines > from joining. I think you missed his point Andrew. He is saying that he was able to download work on two machines that couldn't possibly complete the WU by the deadline. BOINC shouldn't have allowed this to happen. JM7 is pretty savvy to the workings of BOINC. > We should prepare for queries on how to transfer models to a new machine! This is not currently possible with BOINC. Any machine returning a WU that was not downloaded by that machine will not receive credit. <a><img src="http://boinc.mundayweb.com/cpdn/stats.php?userID=18"></a> |
Send message Joined: 5 Aug 04 Posts: 63 Credit: 21,399,117 RAC: 0 |
Navigator, <i>Only</i> 650 hours!? My fastest PC is in the 750 range. Andrew>they will upgrade How many Intel shares do you own? :) Cheers, PeterV. <IMG SRC="http://boinc.mundayweb.com/cpdn/stats.php?userID=4&trans=off"> |
Send message Joined: 17 Aug 04 Posts: 753 Credit: 9,804,700 RAC: 0 |
> I think you missed his point Andrew. He is saying that he was able to download > work on two machines that couldn't possibly complete the WU by the deadline. > BOINC shouldn't have allowed this to happen. > Sorry, I conflated two points and should have explained my thinking! The deadline is purely artificial, as I understand it. The word at the CPDN open day was that the project would still support the current model for the foreseeable future, so the results should still be valid. I agree that BOINC should be automatically excluding machines that cannot complete within the deadline, but my comment was really meant as a suggestion that the project should be generous in admitting machines that do seem underpowered to the rest of us. |
Send message Joined: 5 Aug 04 Posts: 907 Credit: 299,864 RAC: 0 |
well Andrew H has it right, I did have the tag in the config, so Pentium 90's say, wouldn't be able to download as they wouldn't finish even with our ~1 year report deadline. But for the "education aspect" of the project, i.e. schools that may not have very fast machines but would like to run the model, I decided to take off this check. |
Send message Joined: 5 Aug 04 Posts: 172 Credit: 4,023,611 RAC: 0 |
> well Andrew H has it right, I did have the tag in the config, so Pentium 90's > say, wouldn't be able to download as they wouldn't finish even with our ~1 > year report deadline. But for the "education aspect" of the project, i.e. > schools that may not have very fast machines but would like to run the model, > I decided to take off this check. > OK. Since this was done intentionally, it is not a problem, but I don't think I have quite that much patience. With sharing with other projects, I expect that those WUs would not be returned this decade. I am going to keep running them for a while to test new BOINC code, but eventually I would like to ditch WUs 14775 and 14597. Could you mark these to be sent out again? jm7 |
Send message Joined: 7 Aug 04 Posts: 187 Credit: 44,163 RAC: 0 |
> but eventually I would like to ditch > WUs 14775 and 14597. Could you mark these to be sent out again? Oh, can you do that? In that case, I had just started a new WU and Carl mentioned resetting to get the release code if you weren't too far along, so I abandoned it. It was WU 17018. Also, looking at my results, I see a WU I was never sent. (hmm... hasn't this been happening over at Predictor as well?) Apparently I was sent a WU today, then 4 minutes later sent another. Looking at my event log after installing 4.05 and resetting the project, it's clear to see that there was only a single WU downloaded to my system. The log doesn't have enough in it to be reset yet, so I know it isn't just a matter of no longer having record of it. Anyway, the mystery WU is 24481. This is the one that supposedly downloaded first, but doesn't show up on my event log. <a><img src="http://boinc.mundayweb.com/cpdn/stats.php?userID=18"></a> |
Send message Joined: 5 Aug 04 Posts: 390 Credit: 2,475,242 RAC: 0 |
> > but eventually I would like to ditch > > WUs 14775 and 14597. Could you mark these to be sent out again? > > Oh, can you do that? > > In that case, I had just started a new WU and Carl mentioned resetting to get > the release code if you weren't too far along, so I abandoned it. It was WU > 17018. > AFAIK there is an automatic re-scheduling of WUs - if they are stop trickling for some period...? |
Send message Joined: 5 Aug 04 Posts: 907 Credit: 299,864 RAC: 0 |
> AFAIK there is an automatic re-scheduling of WUs - if they are stop trickling > for some period...? actually not right now, in this 7-dimensional parameter space we're using on hadsm3 we can just generate more workunits at will. |
Send message Joined: 17 Aug 04 Posts: 753 Credit: 9,804,700 RAC: 0 |
If the WU is reported to the server as having been terminated (eg crashed) then it is resent, though (ie normal BOINC behaviour) unless Carl has changed something recently. This leaves WUs that are in limbo because the server still has them flagged as in progress, and I guess that the server would resend these once past the deadline. Don't know if the server flag is changed if we use 'Reset project', though. I was holding off a day or two to see if we get to 4.07 by then :-)). |
Send message Joined: 26 Aug 04 Posts: 1 Credit: 43,892 RAC: 0 |
> Welcome on board, Navigator! Yes, all CPDN WU takes 3-6 weeks to complete > depending on your computer. For your Athlon 3200+, you may expect 3-4 weeks if > your run your computer 24/7. > > Hopefully, your will receive credits at the completion of every 10802 time > steps (= 1 trickle) (or appr. at every 8-10 hours of processing). > > Can I bother y'all for a followup question? Does BOINC queue things up based on the deadlines? For example, say I had 4 seti units in place with a deadline of like 8/11 and it starts crunching the first. Then if I get something from Predictor that has a 8/9 deadline, will it finish the first Seti unit and then do the predictor ones with the shorter deadline, and then go back to Seti? My concern is that the year off deadline for CPDN would mean that pretty much everthing would get calculated first because of the shorter deadline, and then once I got really close to the year-off CPDN deadline it would start crunching it... but too late to finish it in time? Advice appreciated. Chris |
Send message Joined: 5 Aug 04 Posts: 907 Credit: 299,864 RAC: 0 |
actually Andrew is right, I just checked and some crashed results were regenerated by BOINC, so that's pretty nice. |
Send message Joined: 7 Aug 04 Posts: 187 Credit: 44,163 RAC: 0 |
> Can I bother y'all for a followup question? Does BOINC queue things up based > on the deadlines? For example, say I had 4 seti units in place with a > deadline of like 8/11 and it starts crunching the first. Then if I get > something from Predictor that has a 8/9 deadline, will it finish the first > Seti unit and then do the predictor ones with the shorter deadline, and then > go back to Seti? This was the normal mode of operation. > My concern is that the year off deadline for CPDN would mean that pretty much > everthing would get calculated first because of the shorter deadline, and then > once I got really close to the year-off CPDN deadline it would start crunching > it... but too late to finish it in time? The 4.xx client uses pre-emptive scheduling. This allows BOINC to pause a WU that is in progress and switch projects. This should give a more accurate project split based on your resource share. How well this works no one can say until someone misses a CPDN deadline due to running other projects. :) <a><img src="http://boinc.mundayweb.com/cpdn/stats.php?userID=18"></a> |
Send message Joined: 31 Aug 04 Posts: 2 Credit: 0 RAC: 0 |
i recive a 3000 hours WU... i guess i have to reset to get it fix? |
©2024 cpdn.org