boinc on ipod.

Dex
Dex
Joined: 21 Sep 06
Posts: 646
Credit: 4381
RAC: 0

I have some data

I have some data here.

This computer, is the slowest computer I could find that could make short work unit deadlines in the past. 166mhz http://einsteinathome.org/host/646847

This computer, is the slowest I could find, that is active, and getting results, and credit. 199MHZ http://einsteinathome.org/host/103769/tasks

http://einsteinathome.org/host/390166/tasks
That computer. Less than 100MB ram, about 266MHZ, and it completes small work units in about 46,000 seconds... Long WU's are about 10 times the size. So, a computer of that ability, could finish a long WU in about 5.6 days. Which is before current deadlines.

My computer could finish a long work unit in about 2.02 days.

Slower computers aren't usually given large work units, but, that is explained in another forum post. That the slowest 20% of computers, recieve small work units.

d3xt3r.net

Odysseus
Odysseus
Joined: 17 Dec 05
Posts: 372
Credit: 19557274
RAC: 3285

RE: Generally I've found

Message 55753 in response to message 55751

Quote:
Generally I've found you need to be in the 1 GHz ballpark for the longs.


My 400-MHz Macs have been getting nothing but shorts for quite a while, but I don’t see what the problem would be should one of them get a long. I doubt they’d take more than thirty hours or so, certainly less than ten percent of the time from issue to deadline. Unless the project is given a very small resource share, or the system’s dedication is low, there shouldn’t be any difficulty getting tasks like that done on time.

Dex
Dex
Joined: 21 Sep 06
Posts: 646
Credit: 4381
RAC: 0

RE: I'd be interested in

Message 55754 in response to message 55747

Quote:
I'd be interested in where you found the flops data for the ipod CPU that was something I couldn't find anywhere, and there was enough variation between different models (Even excluding the ones with a DSP) that I didn't feel confident in making a SWAG.

From my experience, Floating Point Operations per second, varies very little from -15% of actual CPU MHZ Time. What does vary quite a bit, is Integer Operations. Which I was unable to find. The only information on FLOP data, I was able to find after I read your post, was, most likely an assumuption from a poster on another forum. But, when I made my generalized calcuations on crunching time, I took into account, almost lowest possible circumstances. Assuming, low FLOP ability, Slow Cache and CPU Bus speed, and RAM, Heat and OS, lag. Of course my generalized information, is most likely not going to be exactly accurate. But, until someone, actually attempts to implement this idea, theory is all I can work with. ;)

But, with calculations aside, generally, with information I know, I like to think a iPod could finish work units in time for deadlines. That is taking into account that the iPod is running 24\\7.

I try to look at the bigger picture. Maybe one iPod being used for crunching, may SEEM somewhat useless. But, with network computing, it isn't... It's like this idea, if everyone said, what will my one dollar, help a homeless person? True, one dollar may not help. But unless people take the initiative, to help a seemingly small amount, others will have no one to follow. But, I suppose there are several angles to almost every idea. What will 1 Gigaflop add to 72 Teraflops??? Well, it adds a whole gigaflop. Food for thought I suppose...

Hope everyone is well! :D

d3xt3r.net

Vid Vidmar*
Vid Vidmar*
Joined: 22 Jan 05
Posts: 25
Credit: 191816
RAC: 0

RE: Nope, earlier PIII's

Message 55755 in response to message 55751

Quote:

Nope, earlier PIII's aren't fast enough to draw long results. I know 550 and 600 MHz Katmai aren't. A 700 MHz Coppermine might just squeak in barely.

Generally I've found you need to be in the 1 GHz ballpark for the longs.

Alinator

My 300MHz Celeron @450Mhz (view) does a long wu in ~70h. I don't see a reason why an earlier PIII couldn't do it.

Jim Milks
Jim Milks
Joined: 19 Jun 06
Posts: 116
Credit: 529852
RAC: 0

RE: how long would a 667

Message 55756 in response to message 55748

Quote:
how long would a 667 MHz Coppermine need for a short WU? It would only get shorts, right? I can't see it meet the requirements for not being considered a "slow host", really... It has 256 MB of memory (old SD memory, cheap stuff like the mainboard and controllers) and is running Windows XP (I'm convinced it would do better under Linux but no one else in my family really shares that POV ;-) I could only guess how much time it would need, but I seem to remember installing BOINC on it for benchmarking and getting something around 900 integer, 550 floating point, or so. Could be mistaken there.

The slowest computer I'm running on E@H is a 400 MHz G3 iMac. The iMac has 256 MHz RAM with benchmarks of 251 measured floating point speed and 596 measured integer speed. It finishes a short WU in seven to eight hours. So, from a benchmark point of view, your Coppermine would be plenty fast enough for E@H.

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9352143
RAC: 0

RE: RE: Nope, earlier

Message 55757 in response to message 55755

Quote:
Quote:

Nope, earlier PIII's aren't fast enough to draw long results. I know 550 and 600 MHz Katmai aren't. A 700 MHz Coppermine might just squeak in barely.

Generally I've found you need to be in the 1 GHz ballpark for the longs.

Alinator

My 300MHz Celeron @450Mhz (view) does a long wu in ~70h. I don't see a reason why an earlier PIII couldn't do it.

I'm not saying they can't do the long results within the deadline, they can. In fact early in the S5 run I was getting long results on one of my K6-2/500's, which even though are in the same CPU speed class as my PIII's have a far less capable FPU and don't support SSE.

What I'm saying is these slower hosts generally don't meet the selection criteria for the project side scheduler to send them long results. IIRC, a host has to report a CPCS of ~0.0014 to be eligible for long results (you can find that in the last contact log for a host).

I do find it interesting your OC'ed Celeron manages to just squeak over this limit, but you can see that my 550 Katmai and K6-x's don't. I may have to look into this some more and see if I can squeeze a few more FLOP's out of them! ;-)

In any event, I would imagine the scheduler would send a long result to a "slow" host if no shorts were available and the host had adequate BOINC resources to meet the deadline.

Alinator

@ Annika: Based on this, I would almost bet that 667 MHz Coppermine would be fast enough to be eligible for all results.

Vid, it looks like the reason you can get them is the "big" numbers your OC'ed Celery can put up for IOPS. That's about 700-800 Meg more than my Katmai shows in it's current configuration (which has been pretty stable BM wise for months).

Annika
Annika
Joined: 8 Aug 06
Posts: 720
Credit: 494410
RAC: 0

So, it seems that this

So, it seems that this computer could get long results. The actual benchmarks are a bit higher than what I posted earlier; when I checked yesterday it was 580 float/1000 integer. One question: Do long WUs take up more memory? I admit I never bothered to compare the two, since my only host to ever get long WUs (my little laptop is easily fast enough for long ones but always draws short ones in the lottery) has 1 GB of memory so I never noticed a difference. But with only 1/4. would the difference be noticeable? I hope not, because people actually try to work with the thing ;-)

Dex
Dex
Joined: 21 Sep 06
Posts: 646
Credit: 4381
RAC: 0

Long WU's take a bit more

Long WU's take a bit more memory, but very little. With 1GB RAM, you won't notice any difference. It may use only more virtual memory if you have a longer Write to Disk Interval.

d3xt3r.net

Dex
Dex
Joined: 21 Sep 06
Posts: 646
Credit: 4381
RAC: 0

RE: Long WU's take a bit

Message 55760 in response to message 55759

Quote:
Long WU's take a bit more memory, but very little. With 1GB RAM, you won't notice any difference. It may use only more virtual memory if you have a longer Write to Disk Interval.

Unless I am wrong. I have not done to much testing about that issue. SO I am not completely sure.

d3xt3r.net

Lt. Cmdr. Daze
Lt. Cmdr. Daze
Joined: 19 Apr 06
Posts: 756
Credit: 82361
RAC: 0

RE: IIRC, a host has to

Message 55761 in response to message 55757

Quote:
IIRC, a host has to report a CPCS of ~0.0014 to be eligible for long results (you can find that in the last contact log for a host).

Has anyone a clue how CPCS is determined? I have a 2GHZ laptop, but since I'm running diskless its CPCS has reduced (currently 0.0011). So it can't be solely depending on CPU, but also on free HD space...

Regards,
Bert

Somnio ergo sum

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.