Message boards : Number crunching : Why not all cores!?
Message board moderation
Author | Message |
---|---|
Send message Joined: 25 Apr 07 Posts: 8 Credit: 1,900,808 RAC: 0 |
Since Climate Prediction takes so long why not give us the option of using all or the number of cores we want to use for this project. |
Send message Joined: 7 Aug 04 Posts: 2187 Credit: 64,822,615 RAC: 5,275 |
See my response to your GPU question for IT staffing issues. Although, the original had*** models are multi-threaded, the work has not been done to try this in the BOINC environment. It was proposed by the person heading the FAMOUS model project, but fell by the wayside due to other priorities and IT staffing shortages. It may come back if really long models or really complex models are planned, and the IT staff is stable. But there are no short term plans to my knowledge. |
Send message Joined: 21 Nov 04 Posts: 6 Credit: 902,269 RAC: 0 |
I have a six-core machine, so I just run five concurrent climate models (reserving one core for my general use). Chuck |
Send message Joined: 3 Oct 06 Posts: 43 Credit: 8,017,057 RAC: 0 |
Yes, but the advantage to a multi-threaded application would be that a single model is finished much faster. And maybe there would be less contention for resources, resulting in a faster completion time? This would benefit research giving useful results faster. I'm trying to say a multi-threaded application might be useful to the researchers, regardless of the feasibility. |
Send message Joined: 16 Jan 10 Posts: 1084 Credit: 7,827,799 RAC: 5,038 |
I'm trying to say a multi-threaded application might be useful to the researchers, regardless of the feasibility.The key word there is 'might'. CPDN is good at creating large collections of models, so the project needs technologies that bring forward the availability of an ensemble of models rather than any particular model. It could be that making the models quicker helps this, but reducing the high failure rate could be a lower-risk way of getting an earlier 'complete' ensemble (from the statistics of which some scientific decision could be made). Both increased speed and improved reliability also improve the user's experience of CPDN/BOINC, which is another benefit. The project has responded to the oft-stated preference for shorter models (which is not necessarily the same thing as quicker models), and generic BOINC/CPDN faults are eliminated from time to time. So, current server problems notwithstanding, the project is inching in the right direction. |
©2024 cpdn.org