Which PC for which project?
16 Aug 2006 13:25:14 UTC
Topic 191689
(moderation:
Lots of people talk about certain PCs being better for certain projects, due to their architecture, speed, OS or whatever. Now I'm wondering if you could give me a bit of advice on how to figure that one out. I know I could just try it out but I'd really appreciate it if someone would share their experience with me.
The thing is, I will be running BOINC on two computers any time soon and I'm wondering how to assign the proejects to them- I mean, I might run one on one PC and one on the other, or both projects on both PCs... atm I'm doing SETI and E@H with equal resource shares. I wouldn't mind doing a bit more of E@H (so I'm flexible about the percentages) but I definitely want to keep running those two projects. Also, I consider maybe joining a third project but I'm not sure about that and I also have no idea which project might be suitable.
One PC (my current one) is an Intel Celeron M laptop running at 1.3 GHz. Not that terrible fast, but it says in my profile I can crunch up to 32 WUs on it. The benchmarks actually look rather impressive but that is probably due to what it says in the BOINC wiki about Celerons being overvalued in the benchmarks. This comp has 496 mb of memory and what might also be important is that the gfx card just sucks- sorry about that but it really belongs to the very lowtech components out there. In E@H this causes no problems as long as I switch off the screensaver but this comp is definitely not up to any projects requiring complicated graphics, like CP or so. As soon as I have the new box I'm going to have this baby run under Debian Linux, just in case OS is also that important.
The other comp will have an AMD Athlon 3500+ CPU, 1 Gig of RAM, a decent gfx card and all the other components also optimized for a bit more computing power. As I don't have it yet I can't, of course, give you any benchmark results, but wasn't there sth about AMD CPUs doing generally well in E@H? This box will, at least for the time being, run under Windows XP.
Well, I hope anyone has some advice for me, because with the variety of projects and computers out there this can be a bit confusing. Thanks a lot in advance!
Annika
Which PC for which project?
)
On Einstein, as of now, you are already using current optimized aps (was beta a week ago) without doing anything special. On SETI, with a little installation effort, you can raise your production by using Simon's aps (see the sticky new optimized ap thread over on SETI).
As to advice, I suggest you set each machine 50/50, watch the project comparative results, and shift resource share toward the project with comparative advantage on each machine. That is what I do.
Generalizations about Celerons and companies are pretty broad. Intel has sold everything from dogs to Q-ships under the Celeron brand over the years. Talking about Prescott, Banias, and Conroe all as "Intel" covers some pretty different behaviors.
Even once you get the answer, it may change with new aps. For a while my Coppermines were dogs on SETI vs. Einstein compared to my Banias and my Gallatin, so I set the Coppermines to 90% Einstein, the Banias to 90% SETI, and the Gallatin 50/50. But with the current set of aps the playing field among my little set is actually pretty level.
As I said, it is a Celeron
)
As I said, it is a Celeron M... I think that would be a downgraded Centrino. Thanks for your advice about trying to find the project which performs better on a given PC, but how exactly do I figure out that "comparative advantage"? Sorry if this sounds nooblike but I'm really not sure... And another question: Is it really possible to use different resource shares on each PC? If so, how do you set that? As I've never yet crunched with more than one machine I don't know... I thought those settings applied to all computers in one acc...
RE: And another question:
)
"General preferences" allow you to "Add separate preferences" for home, school and work. Therefore, you can define "where" each host is "located" and vary preferences that way (up to 3).
Okay, that never occured to
)
Okay, that never occured to me but it sounds logical :-) thx a lot.
RE: but how exactly do I
)
If you set resource share to 50/50, you can just watch to see which project gets more credit, and by how much. If you set it to some other ratio, just look at the way credit accrued differs from that ratio.
If you are patient, and like an exponentially decaying weighted average, you can just watch RAC. If you wait a month, it will give you a pretty good idea of comparative advantage. If you are impatient, you can plot RAC and try to guess where it is headed, or plot the un-weighted data. Your reward is a faster answer. Your punishment is more noise in the data--a better chance of getting it wrong.
The common thread in all of these is that if you assign resource at 50/50, but the credit rolls in at 65/35 on one machine, and 55/45 on another, you'll get most bang for the buck by raising the resource share for second project on the machine where it has comparative advantage. In this example, while the project getting 45 on machine 2 is still behind the first project, it costs less in forgone credit on the leading project to give the lagging project resource on that machine than on the other. If credit were your only goal, in this case you'd give the leading project 100% share on both machines. But if you prefer to support both, you should devote most of the support on the machine where you get a better trade--just leaving enough on the other to monitor for change (new aps... WU's with different behavior...).
Okay, that sounds really
)
Okay, that sounds really useful, thanks a lot! And no, credit is definitely not my only goal... I like a bit of competition but I would certainly not have joined here if I wasn't fascinated by the science. I could just get myself a computer game if it was all about owning other people ;-) But even so I want to do things efficiently... I mean, it helps the projects aswell if we're not wasting our computing power... and besides, being an informatics student I'm just interested in how all this works and how to make it work even better. :-)
I just noticed there is an
)
I just noticed there is an easier way to find out, btw... because in your account there is a possibility under "your computers" to see how much credit this very machine has gotten so far... so all you'd have to do would be switch the comp on on 50% Einstein 50% SETI, let it crunch for a couple days and then check where it got more credit... of course you'd have to be careful not tp determine this when you've lots of pending credit on one project but apart from that it seems perfectly okay to me, or am I mistaken here?
RE: I just noticed there is
)
It takes about 2-3 weeks before your RAC become stable. So, I would suggest running it for about 3 weeks at 50/50, and then make a determination. Both projects have different lengths of processing times. Einstein has a few physicially different sized units, SETI are all the same size, but the different angle ranges cause for different processing times. So until you get a good share of different results to crunch, it's not going to be very deterministic.
Is there any advantage to
)
Is there any advantage to running a P4 in HT mode on Einstein? I only run two projects on the P4 and that's Einstein and Rosetta. Einstein has a 1/3 share of the computer time and Rosetta has a 2/3 share. Currently running in HT mode at the moment.
I never had a P4... but I
)
I never had a P4... but I read in some other threads that a HT processor like a dual core would be able to process 2 WUs at once, is that true?