Message boards :
Number crunching :
efficiency
Message board moderation
Author | Message |
---|---|
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
Is it more efficient to have three computers (virtually identical) running one project each or running three projects on the three computers with 1/3 resource share? Thanks. |
eberndl Send message Joined: 12 Oct 01 Posts: 539 Credit: 619,111 RAC: 3 |
i would say 3 on 1/3 resource share. If One project goes down, all three will still have units to crunch, but if computer A only does SETI and it shuts down for a week, that's a week waster for that computer. Also you CPIDs would never line up (for stats purposes) if there is no link between your various projects. |
Ziran Send message Joined: 19 Mar 03 Posts: 32 Credit: 721 RAC: 0 |
That depends on how you define efficient. I'd say that that to have three computers (virtually identical) running one project each would be more efficient in theory, since the computers don't have to switch between projects. That is until one of the computers run out of work. To safe guard against that you would have to have a bigger cache. If it's HT computers, then from what i have heard it would be more efficient to only run one project because WU's from the same project seams to coexist better in the L2 cache. The turn around time would be smaller on a 1 project computer. from a project stand point: one computer would mean less connections to the project. So i would say, the most efficient is nether of the alternatives. The most efficient way would probably be to set the resource share to >1000 for main project and 1 for the other two. |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
By efficient, I meant the highest possible production, not taking into account the possibility (inevitability) that the BOINC servers will go down. With a larger cache, I shouldn't miss work. I'm thinking to detach from all but the project assigned to each computer. When the projects crash, I'll have cached work and when cached work is about to run out, I can attach to another project. Does that sound feasible? |
jrmm22 Send message Joined: 30 Jan 04 Posts: 353 Credit: 24,536,157 RAC: 0 |
By efficient, I meant the highest possible production, not taking into account the possibility (inevitability) that the BOINC servers will go down. I think that will give you very good results. I would recommend on using optimized apps (where applicable). and optimized BOINC CC. However, if there's anything wrong with a PC, no work will be done in that project. If you can live with that, then go ahead. I'd do that myself, If I had 'n' PCs available and they were all about the same. |
John McLeod VII Send message Joined: 15 Jul 99 Posts: 24806 Credit: 790,712 RAC: 0 |
By efficient, I meant the highest possible production, not taking into account the possibility (inevitability) that the BOINC servers will go down. Sounds like more hands on management than I want to do. But if it works, shy not. If you have a HT machine, you are better off with two different projects on it as they will use the different portions of the chip in different ways and the flow will be slightly better if they are both running at the same time. BOINC WIKI |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
I already run "just" one project on three of my slowest machines. I use the optimized SETI client and don't think anyone has done an optimized client for Predictor or Einstein. I'm not sure they have "open" code. I ran optimized 4.45 manager on a couple of machines and didn't see that it made any difference and didn't really expect it to. I'm using someone elses modified 4.45, so I have the "no new work" status message displayed. |
Pascal, K G Send message Joined: 3 Apr 99 Posts: 2343 Credit: 150,491 RAC: 0 |
I already run "just" one project on three of my slowest machines. Optimized Manager does nothing to help the processing speed of a WU the optimized Application is the one that processes the WU.... Semper Eadem So long Paul, it has been a hell of a ride. Park your ego's, fire up the computers, Science YES, Credits No. |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
Exactly. That's why I'm not bothering to use an "optimized" manager now. |
jrmm22 Send message Joined: 30 Jan 04 Posts: 353 Credit: 24,536,157 RAC: 0 |
I already run "just" one project on three of my slowest machines. No, but helps to get you better credit (small chance, but hey, its something...) |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
I'd rather have credit for all the WUs I crunch that go South, than to have extra from an "optimized" manager. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
In theory, a random selection of instructions in mixed threads should lead to the "best" productivity on a HT P4 chip. So, I would lean towards having the machine run at least two different projects, over time if one application is blocked there is a good chance that the other thread will be doing something completely different ... therefor, better throughput... Have I tried to clock it? No, then again, I am the guy that says that 4.45 works too ... so what do *I* know ... :) |
j2satx Send message Joined: 2 Oct 02 Posts: 404 Credit: 196,758 RAC: 0 |
I'm all "single-threaded". My dual PIII is slated for my computer museum to be replaced by an A64. My only other "Intel" machine is my sloooooow laptop which I really don't crunch on anymore.............just run BOINC Manager to manage the remotes, so I don't have to pcAnywhere to them. Thanks...........I think 4.45 works too......just doesn't give me instant gratification. |
Kajunfisher Send message Joined: 29 Mar 05 Posts: 1407 Credit: 126,476 RAC: 0 |
just so you know, there are p3's, p4's, and AMD64's that make my 2 year old laptop look like it's sitting still, but still I crunch! btw, running 4.45 (thanks JM7!) |
W-K 666 Send message Joined: 18 May 99 Posts: 19062 Credit: 40,757,560 RAC: 67 |
If you have a HT machine, you are better off with two different projects on it as they will use the different portions of the chip in different ways and the flow will be slightly better if they are both running at the same time. I have not had chance to try it on my sons HT machine but on my Dual P3 I find running one Einstein unit and one Seti unit at the same time increases the productivity on seti by about 15%. The time to process two seti units at same time is average of 8h:20m, but with one einstein and one seti, seti is processed in 7h:10m averaged over run of ten seti units after I noticed this seemed to be happening. Andy |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.