Welcome to MilkyWay@home

Posts by PinkPenguin

1) Message boards : Number crunching : Nbody -INF - Validate Error (Message 57005)
Posted 23 Jan 2013 by Profile PinkPenguin
Post:
Can anyone tell me what the following STD_ERR outputs mean:

a) <search_likelihood>-inf</search_likelihood>
b) <search_likelihood>-1.#INF00000000000</search_likelihood>

Seems like a) produces a "Validate Error" and b) produces "Completed and validated"...

I note that a) seems to be frequent on Linux and Mac OS X machines and b) happens on Windows (at least for the WUs I have examined). This is for the nbody v1.04 app on all machines.

Here is one example with both Linux / Mac and Windows results:
http://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=298948800
2) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 40686)
Posted 26 Jun 2010 by Profile PinkPenguin
Post:
Sorry I have been away for a while. I have run a few v0.31 WUs on the MAC and they seem to be OK even when cross-checked with a second WU.

Looks like the problem been resolved (as already stated in the News thread) - many thanks Travis.
3) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 40060)
Posted 30 May 2010 by Profile PinkPenguin
Post:
In case it is may help the work unit which didn't work for the MAC was run against another 3 tasks all of which completed:

[urlhttp://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=112838646[/url]

The two ATI 4700/48000 v0.23 (Gipsel) apps ran on Windows 7 64-bit and Windows XP 32-bit tasks were validated and given credit while both the Gipsel v0.21 optimised app and the OS X 64-bit v0.29 were marked as invalid.

This is the stderr output for the MAC OS X run:
<core_client_version>6.10.56</core_client_version>
<![CDATA[
<stderr_txt>
<background_integral> 0.00190678182272625797 </background_integral>
<stream_integrals> 322.33718710180266953103 0.73352179106380688278 2409.85511512679568113526 </stream_integrals>
<background_only_likelihood> -3.32514159191908165170 </background_only_likelihood>
<stream_only_likelihood> -4.68067977848977445632 -13.80246848474236287529 -4.69941102303296975862 </stream_only_likelihood>
<search_likelihood> -3.16812541245001977330 </search_likelihood>
<search_application> stock_osx_x86_64: 0.29 double </search_application>
01:53:32 (38211): called boinc_finish

</stderr_txt>
]]>


Both ATI runs which were considered valid seem to ignore input parameters from app_info.xml but I suspect this does not mean anything...

Any results returned by the OS X v0.29 app may not be reliable... as it validates only when it is the only task in the run (i.e. no cross-checking is done).
4) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 40057)
Posted 30 May 2010 by Profile PinkPenguin
Post:
Here is the update - looks like the same problem as before. I suspect the server configuration is OK the problem is the validity/usefulness of results from MACs:

1. If the WUs run by themselves (i.e. there is no quorum) they validate and get credits.
2. If the WUs run to validate "inconclusive" results from other WUs they do not succeed in resolving the issue and a fourth WU has to be sent out. This always results in the MAC WU being invalidated.

The situation suggests that either MAC WUs are right and the rest of the world is wrong or, more unthinkably, MACs are wrong and the rest of the world is right.... ;)
5) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 40034)
Posted 28 May 2010 by Profile PinkPenguin
Post:
I think we need to be patient for a little longer until the software versions change for MAC OS X. Certainly the v0.29 app fails every time it is validated by another app (whether ATI or CUDA or CPU).


I should have updated the server so it doesn't do that validation for the time being... I'll check to see what isn't working.


Thanks Travis - just saw your post and will download a couple of WUs this weekend see how they go and let you know.
6) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39875)
Posted 21 May 2010 by Profile PinkPenguin
Post:
I think we need to be patient for a little longer until the software versions change for MAC OS X. Certainly the v0.29 app fails every time it is validated by another app (whether ATI or CUDA or CPU).
7) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39735)
Posted 15 May 2010 by Profile PinkPenguin
Post:
What I thought I've done is temporarily not validating the OS X applications server side -- they all should be awarded credit. Part of the reason they weren't validating correctly in some cases is that they were using the new summation method in the new version of the application.

I'm pretty much just waiting for Anthony to recompile the new CUDA applications, and for Cluster Physik to do the ATI ones then we'll be make the swap over to that new application, so all this won't be an issue.

It's also been finals week so things have been pretty crazy around here, sorry for not responding sooner!


That's OK I figured you would have lot to do. I'll probably be back next weekend and see how things are coming along.

Thanks for the reply.
8) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39685)
Posted 13 May 2010 by Profile PinkPenguin
Post:
@Paris - yep, I think you're right... can't see how the other boxes could've got it right !

Anyhow I have to switch to Rosetta for a week and then I'll be back to see if there's been any developments.... meantime I'll run through the rest of the queue in the hope that the rest of the world admits it's mistake... ;)
9) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39666)
Posted 13 May 2010 by Profile PinkPenguin
Post:
It seems to happen on all WUs which require a consensus. The following WU didn't agree with an ATI result and is waiting on a Cuda result in order to decide which is valid (given past results my bet is on the ATI card.... curses, foiled again!):

http://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=104606076

In this case both wingmen are Windows and the ATI is a Windows 7 64-bit system running MW v0.23 and the Cuda23 is a Windows Vista 32-bit system running v0.24.... Mine hasn't changed... though it is looking at me reproachfully.

... Not sure how "consensus" works on the MW system - it looks like every WU unit goes out once and, presumably on the basis of some criteria in the result (i.e. "found something interesting", the same WU is sent to another PC for verification. If the second PC agrees with the first the WU is OK if not a third WU is sent out and decides which of the first two is the winner.... (if all three disagree Travis has found his dumb alien.... but I ain't seen that yet... ;)

The problem is that when my MAC sends in a single response and there is no second WU the WU is validated and credits are awarded. Every time the MAC gets a WU which is also run by other PCs the MAC is invariably (as far as I can see) wrong.... this means that single WU results (which are not checked against a second result because they don't find anything worth checking) are probably also wrong!

This makes the results returned my Macs on the v0.29 app unreliable... even when they are awarded credits (not that I'm complaining I don't mind being awarded credits for being wrong.... I usually am).

10) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39640)
Posted 12 May 2010 by Profile PinkPenguin
Post:
[edit]Sorry Travis.... probably driving you up the wall with this one...[/edit]

Although some WUs go through OK it looks like I am still getting invalid WUs with non clear idea of why as they seem to complete normally. Seems a bit chancy whether the WU comes in valid or not. This example came in this morning:

http://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=104342224

This is against two ATI Cards on 64-bit Windows 7 machines:
1. CAL ATI Radeon HD 4700/4800 (RV740/RV770) (512MB) driver: 1.4.556
2. CAL ATI Radeon HD5800 series (Cypress) (1024MB) driver: 1.4.556

My machine is (hostid=111972): Mac Book Pro 2.8Ghz (dual-core) with Mac OS X 10.6.3 and BOINC Client 6.10.51. It is running the 64-bit v0.29 MW app with the following output:

<search_application> stock_osx_x86_64: 0.29 double </search_application>

... Do Macs have problems with sums :-?
11) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39607)
Posted 11 May 2010 by Profile PinkPenguin
Post:
Been updated, let me know if you're getting credit.


One came in just now - got credit OK.

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=119291520

App. Version is: Milkyway v0.29 - was there a mod to the app ?
12) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39601)
Posted 11 May 2010 by Profile PinkPenguin
Post:
Should have another fix in place sometime tonight. An I'll post an update when I do it :)


Best of luck and thanks.... ;)
13) Message boards : Number crunching : Big issue : Milkiway doesn't work on Mac OS X. (Message 39590)
Posted 10 May 2010 by Profile PinkPenguin
Post:
@Paris: If it's any consolation to you I have one WU which ended normally and has been marked invalid and three others which look like they will go that way as they don't agree with the other result on the same WU... ;)

I am still running WUs as I hope this will help Travis fix the problem.... The alternative explanation is that Macs can't do sums... :-o

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=111972

Here are the version details for this machine:
Mac OS X 10.6.3 (Core Duo 2.8GHz), BOINC Client v6.10.51 and Milkyway v0.29 (am also running Collatz on Cuda v3.0.14 driver on the NVIDIA 9600M GT but this should have nothing to do with it.... ;)




©2024 Astroinformatics Group