Welcome to the MacNN Forums.

ianxwin · Apr 19, 2002, 06:28 AM

I'm sure this must have been discussed many times. I apologise if this is the case.
Is it possible to run multiple different clients under OSX, for instance the folding client, the folding at home, and seti?
If it is possible, are there any efficiency benefits?
I've just set up folding@home for macnn, taking off seti which i was crunching for another group, and would like to keep a reduced interest in keeping seti going.
Ian

Arkham_c · Apr 19, 2002, 09:50 AM

You can run as many as you like, but for the most part you only slow them down by running several at once. If you have dual processors, some clients will benefit from two concurrent instances, but it depends on the project.

The_Equivocator · Apr 19, 2002, 10:05 AM

As you would expect, if you run two different clients on a single processor system, chances are that each client will run approximately half as fast as it would if it were running by itself. On average, my one distributed client nabs over 90% of my cpu when I am not doing anything. If I ran SETI and Ubero, for example, they would each only take 45% of the CPU.

My overall productivity would remain about the same, but I would only turn in half as many units for each project.

Shaktai · Apr 19, 2002, 12:20 PM

Your biggest concern would be your level 2 cache. If you were running on a single processor system and only had 256k of level two cache, you may run into problems with overloading the cache which could seriously slow you down. If you have at least 512 k L2 cache, or a multi-processor syseem, you would probably be okay.

So it depends a little on your machine. The only performance advantage would come from if one client was unable to to connect to its server to download additional work units, as sometimes happens, then the other client would keep working and and would use all the available CPU cycles.

ianxwin · Apr 19, 2002, 12:47 PM

Interesting - thanks for the input.
With the clients being intensly computational, your comments agree with what i would assume. However sometimes processes do not efficiently use all the resources they grab. I put seti up allongside folding@home, and noticed that folding hung onto about 80% of the resources and gave seti about 18%.
When I have more time I must do some timing tests - so far I've not even produced one unit for folding (80% through) - so its early days to get ambitious!
Incidentally I run a 400MHz cube and a 500MHz ibook - with a plethora of Intel systems at work - mostly running BSD.
Ian

SkiBikeSki · Apr 19, 2002, 05:03 PM

I've run serveral different projects concurrently with my single CPU G4. I just have to start each client with the right nice value. EG

<BLOCKQUOTE>code:<HR><pre>./setiathome -nice -20[/code]

that starts SETI and makes it the nicest possible. 20 might not be the right value for all projects, but with experimentation you can find a good balance. I found that dnetc does use nice, but it's own priority scheme; 2 is a good value. Then base the other projects nice value accordingly.

reader50 · Apr 19, 2002, 10:11 PM

In theory, you should be able to come out ahead by running different clients at the same time. The reasoning is that each client will not make use of all the CPU's resources. For example, RC5 might use Altivec and the floating point unit, while Ubero might mostly use the integer units.

So, depending on available cache size, CPU design, etc, you could probably get more units from running two "counter-balanced" clients.

In practice, it would take some rather involved benchmarking to determine which clients work together the best, and which make identical demands on the CPU's resources. To my knowledge, no one has attempted such a benchmarking project on the Mac.

Michael A. Peters · Apr 20, 2002, 01:43 PM

Originally posted by reader50:
So, depending on available cache size, CPU design, etc, you could probably get more units from running two "counter-balanced" clients.

That wasn't my experience running both folding @ seti on a 900 Athlon.

What happened is they both took about 50% of cpu (as expected) flip-flopping as to which took more, and my production for both projects was cut exactly in half for that machine- except when folding was down, I would get the full benefit of seti.

I run a local seti proxy so seti being down wasn't ever an issue.

mikkyo · Apr 22, 2002, 03:57 PM

I run a couple DC clients on the faster machines.
I have tried as many as 4, but the really good clients always shine best (dnetc).
If you have a dual, there is nothing wrong with running 2 different clients or 2 of the same, except for dnet since it is super smart and knows how to take over both CPUs.

I think a good mix for a G4 is one instance of dnetc cracking RC5 and one of something else you like. Obviously you wont be kicking ass on either project, but you can certainly get a lot of WUs done and contribute to both.

You can also just be a team player and have everything installed on all machines and run whatever client is most needed by the team.
Or the two most needed clients on each machine.

If you like playing with the nice levels, you can have everything running all at once, but keep the less important to you clients down at Nice 20 and have the important ones cranking away at -20.
Using that method, you can run into the cache and VM swapping problems though.
If you have 1.5Gb of RAM, you can pretty much go nutz and just run everything, especially on a Dual.

SkiBikeSki · Apr 23, 2002, 01:51 AM

I've played with -nice and got everything to run at 20% CPU power. I couldn't even tell so much was going on in the background when I used OmniWeb or Entourage or iTunes etc... OS X is great like that!

ianxwin · Apr 24, 2002, 08:58 AM

Great input, thanks.
I am currently crunching folding - and thought I'd add a bit of umph to my cube and ibook. So at work I put folding on the PIII 300MHz. It is acheiving about 8% per day! On my systems a WU seems to take about 8 hours.
I cannot believe a PIII is that bad? - Certainly not one to multitask.
Ian

Shaktai · Apr 24, 2002, 09:18 AM

Originally posted by ianxwin:
Great input, thanks.
I am currently crunching folding - and thought I'd add a bit of umph to my cube and ibook. So at work I put folding on the PIII 300MHz. It is acheiving about 8% per day! On my systems a WU seems to take about 8 hours.
I cannot believe a PIII is that bad? - Certainly not one to multitask.
Ian

With folding, the size of the work units will vary dramatically. The Mac & PC's actually get their work units from different servers. The Mac's will consistently get units that are worth from 1/2 to 2 points. The PC's will get units that can go up to 5 points. Those units will have a 9 day turn around limit. Make certain that you are using the CLI client and not the Screensaver. You might also trying running it in "quiet" mode. Otherwise your unit will not be finished in time.

I think that they used a 400mhz PIII for their benchmarking, so anything less may be too slow. I would recommend that you use the PIII for a different project that does not have "time limits". dFold (distributed folding) would be a good one for that machine.

ianxwin · Apr 24, 2002, 02:48 PM

Thanks for the input Shaktai, I'll switch it over to dfold or seti.
Ian

mikkyo · Apr 24, 2002, 03:13 PM

Originally posted by Shaktai:
The Mac & PC's actually get their work units from different servers. The Mac's will consistently get units that are worth from 1/2 to 2 points. The PC's will get units that can go up to 5 points.

So what is their reason for this? Can we change the server to get the higher point units on our macs? This seems odd. Is there different code being run on the different WUs?

Shaktai · Apr 24, 2002, 08:31 PM

Originally posted by mikkyo:


So what is their reason for this? Can we change the server to get the higher point units on our macs? This seems odd. Is there different code being run on the different WUs?

Technically it was because too many macs were returning units late, and therefore they didn't count.

Personally, I think it was too many Mac users whining about it taking too long, when they were trying to run OS-X on old machines that were just too slow and not supported for it. The benchmark machine that determines the turn around time was a Pentium III 400mhz. Basically anything slower than that, Mac or PC will not qualify for the project. My Macs have no problem with any of the proteins, but the "whiners" I personally think messed it up for Mac users. PC users whined too, but not nearly as much. Any way for now, we are working projects that provide smaller work units that are quicker to process.

The macs are a little slower, because the client software is not as well optimized for them. Improvements in OS-X and better patience from Mac users will in the long run, improve our status and the work we get. It is also important to run the GUI client "hidden" not minimized to the dock, otherwise the Mac's take for ever to process. "hidden" they are as fast as the CLI client.

Keep in mind that the higher point work units, also take a lot longer to process.

[ 04-24-2002: Message edited by: Shaktai ]