|
Questions and Answers :
Macintosh :
Shared Memory, other thread locked and pinned
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
![]() Send message Joined: 9 Jan 07 Posts: 467 Credit: 14,549,176 RAC: 317 |
|
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
Iain, Yes, 5 hours on both roughly ... so, looking good ... by now I would have expected them to die ... I got some other project things going on, so it looks like they will be in the \"hopper\" for now as will have to catch CPU when then can ... though I did not want 3 CPDN to run ... ah well ... luck of the draw I suppose ... Oh, and all three were running at the same time ... though I stil have 12.14 G Free RAM ... now what to do with that space??? :) |
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
Just an update, I have 82 hours on one slab model with 87 to go ... the other I have paused but it has 9:41 on its clock (179) to go ... So, it is looking up ... of course, it could be that there is a trap waiting in the next few hours ... :) Knock wood ... |
![]() ![]() Send message Joined: 29 Sep 04 Posts: 2363 Credit: 14,611,758 RAC: 0 |
That is mighty fast crunching on the Mac! http://climateapps2.oucs.ox.ac.uk/cpdnboinc/result.php?resultid=7405528 I believe the first machine to break the 1sec/timestep barrier on a HADCM model was a Mac that Tolu had in Oxford. (I know yours are HADSMs.) It was rather like the first 4-minute mile. I would want to back up the contents of the BOINC folder (if you\'re not already doing this). If a model did crash and you had work from other projects also running, you could suspend the other CPDN model, set every project to No New Tasks, run the work from every other project down to zero and report it, and then restore the backup. After restoring the backup you\'d wait a day or two to make sure all the other project servers really knew about your reported WUs, then reset every project except CPDN. Then get new work from wherever you wanted. That\'s a way of minimising the loss of crunching time and credits. Anyway, this Mac of yours is really performing. Is it running at stock speed? Cpdn news |
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
The problem is that the backup I am using TimeMachine is not very configurable and I have little use for 10,000 copies of transient data ... so, I have excluded the folder ... The machine is \"stock\" in the sense that I did not alter the configuration and set-up other than to install additional parts. For one, I added memory to fully fill the two riser cards with 16G of all matched memory sticks ... I added 1TB drives which I made into a RAID 5 array using the RAID card from Apple ... Other than that, it is a stock system ... :) But, my understanding is that the memory bus is slightly differnent somehow as that was the question the folks at Micromat asked when TechTool 4 would not boot from the CD ... I think the one machine is doing about 4-5,000 CS per day ... it is hard for me to say exactly as I have been playing with projects too much to be able to isolate the effects easily ... Though, if I can get the second Dell dual Xeon back (my brother said he would ship it this week) I may put both of those machines on just one or two PC only projects for a few days to a week to catch them up, and if that proves interesting I may wait for the daily numbers to \"settle\" and from there will be able to have an idea of what it can do ... I am slowly putting my projects into the resource shares I have been planning as I work a project up to its initial target, then I move it to its target share (12 or 14 to 100) allowing me to run as many as 20 project on the one machine with actual CPU allocation from 1.5% to 12.5% ... Like right now I am concentrating on NQueens and Sztaki which have shares of 200 and 348 while I have other project in their allocation slots now ... for instance CPDN should get half a processor each day for working ... though I may raise that until I can clear the models out ... |
![]() ![]() Send message Joined: 29 Sep 04 Posts: 2363 Credit: 14,611,758 RAC: 0 |
I think your 8-core Mac will be getting more than 5000 credits per day. Cpdn news |
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
I think your 8-core Mac will be getting more than 5000 credits per day. It is likely to be above 4,000 I agree on that. It is kinda hard to say for sure as I am, and have been, moving about quite a bit. Right now I think I have about 9 projects with active and continual tasks on it .. But, I have found at least one project MilkyWay@Home that seems not to be running as fast as it should, but, I have mentioned this on their boards and with luck they will figure out what is the issue ... because of this \'butter-fly\" of mine it will be hard to tell for sure ... The other complication is the cross-project credit thingie ... I better stop here as I usually get into trouble when I go on ... :) |
![]() ![]() Send message Joined: 29 Sep 04 Posts: 2363 Credit: 14,611,758 RAC: 0 |
Nobody will ever accuse you of getting too many credits from CPDN. In some ways CPDN ought to be the cross-project benchmark. When CPDN joined BOINC, as far as I know the credit level was aligned with Seti. The CPDN level has to my knowledge never changed except that credits for the current optimised HADCMs were raised by 20% because they run 20% slower. Or approx those percentages. So in some ways we should be the gold standard. But that will never happen because the technology of different computers and types of WUs from different projects are no longer necessarily comparable. What will emerge in the long term as a cross-project credits system is at the moment anybody\'s guess. Cpdn news |
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
Nobody will ever accuse you of getting too many credits from CPDN. In some ways CPDN ought to be the cross-project benchmark. When CPDN joined BOINC, as far as I know the credit level was aligned with Seti. The CPDN level has to my knowledge never changed except that credits for the current optimised HADCMs were raised by 20% because they run 20% slower. Or approx those percentages. There is a spreadsheet that embodies a concept that I proposed to do this very thing ... it uses computers that are attached to multiple projects to show the \"earnings\" rate ratio between the projects ... I had proposed a somewhat similar system with the use of inter-project linkages to allow the normalization of awards. In a sense, it built on the idea of calibrating computers and then using those to establish the productivity ... in that we do have a variety of computer multiple computers will have to be used and this will be by no means perfect ... The problem is that it will require a cross-project coordination that is still lacking ... one of my other pet \"themes\" that many seem to be tired of hearing about ... In my two year \"vacation\" I would have thought that some progress would have been made on some of the issues I was pointing to back then, yet on my return I find not status quo, but a more hostile and even less coordinated environment ... I see strong evidence that many projects, particularly the new ones are running into the same problems that older projects have already faced and the same questions are asked and answered and no one is writing it down ... not even that nut Paul D. Buck ... what ever happened to that crazy guy ... |
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
Back on topic ... The first of the two slab models completed this morning and though the last few moments of 100% and nothing was seemingly going on was worrisome, all is well with patience. Zipped up the model and uploaded, reported and marked as valid .... So, good work team ... (Go team! Go! Rah! Rah!) and I let the second model off the leash and it is running as we speak and has a couple more hours on its clock (now 11:11 with 173:47:20 to go) ... and as I change my processing priorities as another project has come close to initial target I am working down my work queue for it and starting to build up the work queue for the other project ... All seems to be semi-well in Paul\'s world ... |
![]() ![]() Send message Joined: 13 Jan 06 Posts: 1498 Credit: 15,613,038 RAC: 0 |
Congratulations on the slabs :-) Probably the first Leopard slabs to successfully finish. I'm a volunteer and my views are my own. News and Announcements and FAQ |
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
Slab singular ... the second is in work with about 20 hours on it ... I have it set to run continually so about a week I guess, 160 hours, and maybe I will have two in a row ... then onto the coupled model that will take just a little bit longer ... Then we shall see ... |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
The \"nothing going on\" at the end of model is all of the post-processing of the many data files. It can take 10 minutes or more, depending on processor speed. Interrupt it at this point and you\'ll lose it. Quite a few people have. |
![]() Send message Joined: 31 Aug 04 Posts: 239 Credit: 2,933,299 RAC: 0 |
The \"nothing going on\" at the end of model is all of the post-processing of the many data files. It can take 10 minutes or more, depending on processor speed. It was \"tempting\" to intervene ... but, I called up the graphics and it said POST Processing ... so, I waited ... Anyway, one down ... one more in work and then back to the long model ... THEN, if I am going to stay BOINCing, we shall see about another model ... Still think we should have a setting on the prefs page to say only download one model (though some overlap late phase 3 would be Ok, but, I should not have to do gymnastics to try to keep from getting three models that clutter up my lists for months at a time ... Anyway, that is for later ... |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
There are 2 ways to only get one climate model at a time: 1) This is a \'kind of\' setting for getting only one model. A) Download a model. B) Crash it. C) Repeat this until the server has set your quota-per-day to 1. Now you\'ll only get one model at a time. **************** 2) A) Set your number of processors to 1 BEFORE reaching the point of needing a new model. (Or before doing whatever else you did to start getting them. B) After you have one model, set the Project to \"No new tasks\". C) Reset the number of processors to whatever it was before. (Also possible by having 2 venues, one with the number of processors for \'normal\' use, one with the number of processors set to 1. Switch venues before clicking \"Allow new tasks\".) |
©2024 climateprediction.net