Windows Beta Test App 4.24 available

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686043038
RAC: 586568

RE: RE: Some frequencies

Message 69259 in response to message 69258

Quote:
Quote:

Some frequencies really seem to be more affected than others, I had a series of WUs lately where I had a 50 % failure rate (!), and a single result I investigated contained as many as 100 mismatches between Windows & Linux in the final "toplist" containing the 10000 most promising items.

A) Where do you go to see that kind of stuff?

B) Where do you come up with the time to do that analysis? ;-)

A): Just disable network access when you know you have a "special" workunit, so the resultfile can't be immediately sent to the server. Wait until the resultfile is generated, unzip it, and look into it :-). Resultfile is plain ASCII and can be analysed pretty well. It's not too difficult to re-run a WU on another BOINC installation as well (offline, so you don't sent something to the server, of course!!!)

B) I guess being single atm helps a lot :-). I spent maybe a weekend on this because I was curious what kind of error this was (just a small "epsilon" problem which could be fixed by relaxing the validator or something more complex). I'm an IT professional like you, so sometimes you can't help and just HAVE to find out, I guess.

CU

BRM

Donald A. Tevault
Donald A. Tevault
Joined: 17 Feb 06
Posts: 439
Credit: 73516529
RAC: 0

RE: RE: RE: Are you

Message 69260 in response to message 69254

Quote:
Quote:
Quote:


Are you saying that if a Linux host are paired up with any Windows OS that the Linux result is guaranteed to be declared "invalid"? I could've swore that I had one result validate against Linux some time ago, but I've been paired with nothing but other Windows versions lately.

Edit: It must be something specific with either the Linux or the Windows installation that is causing the issue, because not everything fails. I guess it could also be the frequency range of the result. Don't know. I'm currently watching FalconFly's results for comparisons against my processor. Though he said that he wouldn't be really processing a lot of stuff until this week, this WU of his validated against a Windows XP Pro host...


My Linux box just got 304.31 credits against a Windows box. Never had a validation error on my PII Deschutes running SuSE Linux 10.1, BOINC 5.8.17. Same in SETI and QMC.
Tullio

Hmmm... Yep, I knew there were hosts out there going along fine. Obviously something triggers the problem...but what? Is it perhaps a problem with the WU generator? Pure speculation...and I'm probably talking out of ignorance at this point... LOL

P.S. - I HATE REGULAR EXPRESSIONS!!!!! Been up all night working on something for school. I realize they are "powerful and all", but they are way too nerdy, IMO...

Regualr Expressions are cool, but the the syntax is fairly difficult to get used to.

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282700
RAC: 0

RE: Regualr Expressions

Message 69261 in response to message 69260

Quote:

Regualr Expressions are cool, but the the syntax is fairly difficult to get used to.

Like I said, too nerdy... I'd rather let the ubernerds hack at this kind of thing and just give me some private function to call. Of course, some prospective place of employment would naturally decide they wanted something that was not in the black-box implementation... :sigh:

Try doing a google search for "hate regular expressions" and see how many hits there are... LOL

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282700
RAC: 0

RE: A): Just disable

Message 69262 in response to message 69259

Quote:

A): Just disable network access when you know you have a "special" workunit,

I knew you'd come up with some use for zapping network... ;-)

Quote:
It's not too difficult to re-run a WU on another BOINC installation as well (offline, so you don't sent something to the server, of course!!!)

I'm just a poor boy... I can't go getting other hardware... I might check into VirtualPC 2007 or VMware player...but not until after this class is over. Too much to tackle getting set up otherwise (got to get apache or something like it set up this week).

Quote:
so sometimes you can't help and just HAVE to find out, I guess.

Can relate...

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686043038
RAC: 586568

RE: I'm just a poor

Message 69263 in response to message 69262

Quote:

I'm just a poor boy... I can't go getting other hardware... I might check into VirtualPC 2007 or VMware player...but not until after this class is over. Too much to tackle getting set up otherwise (got to get apache or something like it set up this week).

The cross-platform tests I performed were done on the same hardware, just to make sure. I used my regular Win XP and a Linux Live-DVD (Knoppix in this case). (insert DVD, boot , enjoy Linux). Zero installation effort, very nice.

CU

BRM

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282700
RAC: 0

RE: The cross-platform

Message 69264 in response to message 69263

Quote:

The cross-platform tests I performed were done on the same hardware, just to make sure. I used my regular Win XP and a Linux Live-DVD (Knoppix in this case). (insert DVD, boot , enjoy Linux). Zero installation effort, very nice.

What kind of hardware though? I have a sound card that has no official Linux support (X-Fi XtremeGamer), and a Logitech MX Revolution mouse that may or may not be supported, and since I can't mod x11 stuff when running off of an image file (at least I don't think I can).... ? More concerning would be if my network adapter (Realtek GigaLAN) is supported...?

Donald A. Tevault
Donald A. Tevault
Joined: 17 Feb 06
Posts: 439
Credit: 73516529
RAC: 0

RE: I'm pretty sure that

Message 69265 in response to message 69255

Quote:

I'm pretty sure that the cross-platform validation problem neither has to do with the memory access bug we fixed, nor with a particular machine.

IMHO it is rather a problem of numerical differences between libraries or compilers that is triggered only by certain Workunits. It may be the actual frequency value or the data (files), e.g. a certain type of noise in particular frequency bands.

We're working on it.

BM

This may sound like a silly question, but. . .

If there are differences between the various compilers and math libraries, how do we know which ones will give scientifically accurate results? And, have a lot of us been producing results that are worthless?

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686043038
RAC: 586568

RE: This may sound like a

Message 69266 in response to message 69265

Quote:

This may sound like a silly question, but. . .

If there are differences between the various compilers and math libraries, how do we know which ones will give scientifically accurate results? And, have a lot of us been producing results that are worthless?

Interesting questions indeed. Follow up question: One way to improve validation would be to inject simulated "Pulsar Signals" into the input data and verify that the clients find them. Are there any plans to do that in the future?

CU

BRM

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 1580

RE: RE: This may sound

Message 69267 in response to message 69266

Quote:
Quote:

This may sound like a silly question, but. . .

If there are differences between the various compilers and math libraries, how do we know which ones will give scientifically accurate results? And, have a lot of us been producing results that are worthless?

Interesting questions indeed. Follow up question: One way to improve validation would be to inject simulated "Pulsar Signals" into the input data and verify that the clients find them. Are there any plans to do that in the future?

CU

BRM

This was done with some data at the end of the s4 run.

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282700
RAC: 0

Finished first result with

Finished first result with 4.24. Seems a bit slower, but dunno. No provocation (yet). I'll do that with one of the next group that I get...

Comment viewing options

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