Windows S5R3 App 4.07 available for Beta Test

rbpeake
rbpeake
Joined: 18 Jan 05
Posts: 247
Credit: 219,266,664
RAC: 0

RE: RE: Early indications

Message 73519 in response to message 73518

Quote:
Quote:
Early indications are that adding the EAH_NO_GRAPHICS file increases efficiency of processing the S5R3 work units by approximately 15%.

That is a huge difference. But just what are you comparing?


4.07 with EAH_NO_GRAPHICS versus 4.07 without that file. So it is a fair test versus control situation on the same machine (assuming the work units are similar!), but very preliminary results because the test work unit is not even 20% to completion.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,516
Credit: 455,957,483
RAC: 46,162

RE: 4.07 with

Message 73520 in response to message 73519

Quote:

4.07 with EAH_NO_GRAPHICS versus 4.07 without that file. So it is a fair test versus control situation on the same machine (assuming the work units are similar!), but very preliminary results because the test work unit is not even 20% to completion.

Unlike in S5R2, workunits worth the same credits in S5R3 seem to differ in crunch time considerably. Difference can be as high as 20%, actually! So we would need more results before celebrating :-).

CU

H-B

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282,700
RAC: 0

RE: Unlike in S5R2,

Message 73521 in response to message 73520

Quote:

Unlike in S5R2, workunits worth the same credits in S5R3 seem to differ in crunch time considerably. Difference can be as high as 20%, actually! So we would need more results before celebrating :-).

I would think there'd be some difference in performance, but not 20%... I know that the SETI optimizers have both graphics and no graphics packages (I use the one without graphics), but I don't know how much of a performance difference there is over there... Maybe Alinator would know...

Brian

archae86
archae86
Joined: 6 Dec 05
Posts: 2,823
Credit: 3,300,939,506
RAC: 2,550,417

RE: I see that it probably

Message 73522 in response to message 73506

Quote:
I see that it probably makes more sense to run this Beta if you already got S5R3 work.
BM


I've implemented a scheme to run both S5R2 using beta 4.40 and S5R3 using beta 4.07 on the same machine. No waiting for work.

Procedure:
1. if S5R3 units are already on machine, pause any not yet started, and finish any already begun (remember the incompatible checkpointing).
2. stop boinc
3. assure the enstein project directory has the S5R2_4.40 and the S5R3_4.07 .exe and .pdb files. (the last S5R2 beta, and the first S5R3 beta)
4. create an ap_info.xml file which has separate sections for einstein_S5R2 and einstein_S5R3. To be able to process already downloaded files, assure there are two ap_version sections in each ap, 438 and 440 for S5R2, and 401 and 407 for S5R3.
5. restart

Good things:
1. This scheme will correctly process already downloaded and also partially processed S5R2 results, whether you were running 4.40 by ap_info.xml, or 4.38 from the automatic download.
2. This scheme will let you process both S5R2 and S5R3 new work, as it is currently distributed.
3. This scheme will let you process already downloaded but not yet partially processed S5R3 results.

Limitations and bad things:
0. you can't finish partially processed S5R3 work once you've implemented this ap_info.xml.
1. if you skip step 1 of the procedure, then any S5R3 partially processed on 4.01 will error out.
2. any S5R3 already downloaded is already "branded" as 4.01, so will be returned marked that way, even though 4.07 will actually have processed it. This may degrade the beta value to Bernd--so if he objects we should not do this. I think in that event simply omitting the extra 401 ap_version section would cause the already downloaded S5R3 results to error out and be retransmitted promptly.

archae86
archae86
Joined: 6 Dec 05
Posts: 2,823
Credit: 3,300,939,506
RAC: 2,550,417

I've run 4.07 on 8 sequential

I've run 4.07 on 8 sequential S5R3 results with the following CPU times:

h1_0542.70_S5R2__148_S5R3a 28935.94
h1_0542.70_S5R2__147_S5R3a 29318.39
h1_0542.70_S5R2__146_S5R3a 29424.39
h1_0542.70_S5R2__145_S5R3a 29295.84
h1_0542.70_S5R2__144_S5R3a 29101.19
h1_0542.70_S5R2__143_S5R3a 29194.05
h1_0542.70_S5R2__142_S5R3a 29822.56
h1_0542.70_S5R2__141_S5R3a 29644.75

mean CPU time for these eight results was 29342 seconds.

These were all run on a Q6600 with a 3.24 GHz clock rate. The EAH_NO_GRAPHICS files was present in the BOINC directory.

The same host had previously run results in the same sequence on production 4.10. I don't use the screen saver, nor anything else that gets any graphics from the Einstein ap on my screen:

h1_0542.70_S5R2__155_S5R3a 28829.03
h1_0542.70_S5R2__155_S5R3a 28829.03
h1_0542.70_S5R2__154_S5R3a 28860.92
h1_0542.70_S5R2__153_S5R3a 29082.97
h1_0542.70_S5R2__152_S5R3a 28757.94
h1_0542.70_S5R2__151_S5R3a 28898.08
h1_0542.70_S5R2__150_S5R3a 28780.22
h1_0542.70_S5R2__149_S5R3a 29082.97

mean CPU time for these eight results was 28890 seconds

So the 4.40 results took about 1.6% more time, but given the wide observations of execution time variation for even sequential S5R3 results I'd only say it looks equivalent.

Any hope of 20% improvement from the graphics suppression flag for those of us who don't run the graphics anyway is not supported here.

Stick
Stick
Joined: 24 Feb 05
Posts: 790
Credit: 3,336,604
RAC: 3,554

This one just reported and

This one just reported and validated: http://einsteinathome.org/task/87149420

Stick
Stick
Joined: 24 Feb 05
Posts: 790
Credit: 3,336,604
RAC: 3,554

This one appeared to finish

This one appeared to finish OK but then errored out on a -161 file xfer error: http://einsteinathome.org/task/87166974

googloo
googloo
Joined: 11 Feb 05
Posts: 39
Credit: 10,255,887
RAC: 1,517

This one finished and

This one finished and validated. Credit the same as with 4.01. That is, all S5R3 work units have been issued a credit of 219.42, with a range of CPU time from 66,101.31 to 73,135.44. FWIW, I do not use BOINC as my screen saver.

Mikie Tim T
Mikie Tim T
Joined: 22 Jan 05
Posts: 104
Credit: 210,413,582
RAC: 118,563

I've gotten a few results

I've gotten a few results back on my test host, 710154, all valid so far. Going from 4.40 in SR2 averaging 290 CPU sec/cobblestone, to 4.07 in SR3 resulted in averaging 311 CPU sec/cobblestone, so there's still some work to do to rectify the 7% drop in credit/performance that I've witnessed, even with graphics disabled in the app. That's outside the bounds of what Bernd stated as being normal, and is consistent over several 4.07 results now.

Erik
Erik
Joined: 14 Feb 06
Posts: 2,815
Credit: 2,645,600
RAC: 0

There was a credit adjustment

There was a credit adjustment made months ago, which was fine and dandy. But what project is considered the "standard" that Einstein@home is trying to align crediting with? Is there a consensus amongst all BOINC projects as to a standard?

I believe it should be left well enough alone, and let other projects align themselves up to Einstein if they wish.

Why was an adjustment made coincidental to the S5R3 run?

Comment viewing options

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