Information about the new S5 workunits

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 739076781
RAC: 1263353

RE: Just finished my first

Message 37727 in response to message 37725

Quote:
Just finished my first S5R2 workunit @ 49 hours. Pages and pages of printout about checkpoints etc. Is this to be expected?

Yup, normal. The bulk of the output is a count of skypositions. In case of a crash of the app, this allows to pinpoint the portion of data the app was working on at that moment. Your WU had about 25000 skypositions so roughly 25000 * (5 digits plus a comma plus one space) = a bit over 175kByte produced. Only the last few KB are returned with the result, I think (not sure).
Spread over the two days of your run, it's only a couple of bytes per second, so it's not significant, neither IO-wise nor performance-wise.

EDIT: Arrgg..Annika was faster in replying :-)

CU

BRM

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

Yep, but your explanation was

Yep, but your explanation was more thorough. I didn't think of the bit about sky positions; just noticed all those "checkpoint" messages in my logs aswell and heard about it from others (also from the developers who said it was to be expected).

Dogbytes™
Dogbytes™
Joined: 9 Feb 05
Posts: 320
Credit: 423603
RAC: 0

Here's a crashed work unit

Here's a crashed work unit that remain frozen on my Power Mac G5 for over 2 days.

I let it continue frozen to see if it could unstick itself, but to no avail. Another user using a dual AMD Opteron had no problem completing the work unit successfully.

This is just an FYI.

Dronak
Dronak
Joined: 21 Mar 05
Posts: 28
Credit: 10402879
RAC: 0

RE: RE: Just finished my

Message 37730 in response to message 37725

Quote:
Quote:
Just finished my first S5R2 workunit @ 49 hours. Pages and pages of printout about checkpoints etc. Is this to be expected?

I think this is normal. One of my PCs ned 55 hours..

I just checked on my results, and saw that I had an Einstein unit returned last night that took 84 hours -- over 3 days -- of CPU time. I assume it's an S5R2 unit since I see that string in the result name and the application is einstein_S5R2 4.17. I think my previous units were taking 1-2 days, but this is the first time I can recall seeing a unit take this long. Is this a reasonable amount of time for these units?

Dogbytes™
Dogbytes™
Joined: 9 Feb 05
Posts: 320
Credit: 423603
RAC: 0

RE: RE: RE: Just

Message 37731 in response to message 37730

Quote:
Quote:
Quote:
Just finished my first S5R2 workunit @ 49 hours. Pages and pages of printout about checkpoints etc. Is this to be expected?

I think this is normal. One of my PCs ned 55 hours..

I just checked on my results, and saw that I had an Einstein unit returned last night that took 84 hours -- over 3 days -- of CPU time. I assume it's an S5R2 unit since I see that string in the result name and the application is einstein_S5R2 4.17. I think my previous units were taking 1-2 days, but this is the first time I can recall seeing a unit take this long. Is this a reasonable amount of time for these units?


Mine are talking about 18-20 hours, but the work unit I reported actually froze up...lapse time was just over 9 hours, then it got stuck on stupid.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 739076781
RAC: 1263353

RE: RE: RE: Just

Message 37732 in response to message 37730

Quote:
Quote:
Quote:
Just finished my first S5R2 workunit @ 49 hours. Pages and pages of printout about checkpoints etc. Is this to be expected?

I think this is normal. One of my PCs ned 55 hours..

I just checked on my results, and saw that I had an Einstein unit returned last night that took 84 hours -- over 3 days -- of CPU time. I assume it's an S5R2 unit since I see that string in the result name and the application is einstein_S5R2 4.17. I think my previous units were taking 1-2 days, but this is the first time I can recall seeing a unit take this long. Is this a reasonable amount of time for these units?

Hi!

The S5R2 units are longer than the S5R1 and S5RI units, and they come in different "sizes", yours was a long one.

Still, the performance on your machine is too low. My Pentium III 866 actually is faster, so something is wrong here. Did you use the "screensaver" graphics during the run? This might slow the run down quite a bit, it's better to leave graphics off.

Your PC is an Intel Core Solo, right?

This one should perform 2.5 to 3 times faster than it is doing now.

CU

BRM

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

My Core Duo (shouldn't be

My Core Duo (shouldn't be much of a difference since each WU gets processed by only one core anyway) takes a little less than 50 hours for the big 500+ credit WUs. Very similar clockspeed, too, so that would be a realistic finishing time. I think Cores suffer rather much from lack of optimization with the current app, but not THAT much.

Odysseus
Odysseus
Joined: 17 Dec 05
Posts: 372
Credit: 20627216
RAC: 7249

RE: My Core Duo (shouldn't

Message 37734 in response to message 37733

Quote:
My Core Duo (shouldn't be much of a difference since each WU gets processed by only one core anyway) takes a little less than 50 hours for the big 500+ credit WUs. Very similar clockspeed, too, so that would be a realistic finishing time. I think Cores suffer rather much from lack of optimization with the current app, but not THAT much.


I haven’t seen any 500-credit WUs yet, but I have two C2D-based iMacs that are respectively completing 404-credit tasks in about eighteen hours of CPU time, and 172-credit tasks in about seven and a half.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 739076781
RAC: 1263353

RE: My Core Duo (shouldn't

Message 37735 in response to message 37733

Quote:
My Core Duo (shouldn't be much of a difference since each WU gets processed by only one core anyway) takes a little less than 50 hours for the big 500+ credit WUs. Very similar clockspeed, too, so that would be a realistic finishing time. I think Cores suffer rather much from lack of optimization with the current app, but not THAT much.

Hi Annika!

Will you switch your Core Duo to Linux again for crunching, now that the "signal 11" problem seems to be resolved? It would be interesting to see the direct comparison. I've got a feeling that your host should be quite a bit faster under Linux.

@Dronak: Do you have a Core Duo or Core Solo? The detection mechanism in BOINC doesn't seem to be perfect, e.g. my Core Duo (Mac mini) is listed as a Core Solo. If you ghave a Core Duo you might want want to change the prefences so that 2 CPUs (= cores) can be used for crunching simultaneously.

CU

BRM

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

Might be worth a try. I

Might be worth a try. I thought it wasn't worth the bother with an Intel CPU because I expected the difference to be minimal. But if you think there might be a relevant performance gain I'll try it. Still got some problems to get my ATI gfx card and Ubuntu Feisty Fawn acquainted, and worry about the IPSec network at uni, but maybe I should see it as a challenge ;-)

Comment viewing options

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