New Improved Gravational Wave App - Discussion

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4964
Credit: 18720785871
RAC: 6411638

Yes, test apps is needed per

Yes, test apps is needed per this post by Bernd.

https://einsteinathome.org/goto/comment/223686

You can download whatever app you need directly for incorporating into an app_info by modifying the link address with the proper name of the app.  Bernd provided the direct download link here in this first post.

https://einsteinathome.org/content/cuda-application-o3ashf-search?page=3

As an example, here is the 1.08 app link.

http://einstein2.aei.uni-hannover.de/EinsteinAtHome/download/einstein_O3AS_1.08_x86_64-pc-linux-gnu__GW-cuda

Here is the 1.14 app link.

http://einstein2.aei.uni-hannover.de/EinsteinAtHome/download/einstein_O3AS_1.14_x86_64-pc-linux-gnu__GW-cuda

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6441
Credit: 9571527114
RAC: 8342070

Thank you, Keith.

Thank you, Keith.

The post seems to imply that 1.14 should show up with test apps off and 1.15 with test apps on?

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Eugene Stemple
Eugene Stemple
Joined: 9 Feb 11
Posts: 67
Credit: 374671491
RAC: 543234

Keith & Tom, "The post

Keith & Tom,

"The post seems to imply that 1.14 should show up with test apps off"

Yes, Bernd indicated he "promoted" 1.14 out of testing.  That's why I was expecting that cuda app to get downloaded to me automatically and I wouldn't need any app_info.xml or manually fetching the 1.14 app and could properly turn off "run test apps".   As a "shot in the dark" I downloaded the 1.14 app, put it in the project directory, and marked it as executable.  But, of course, the servers don't know I have it and are, therefore, not sending work for it.  In "theory" I could edit the client_state file to make it known to the servers but there's more to that route (including O3ASHF1b_x.config files, for example) than I am confident to do correctly.  There are workable app_info.xml examples posted elsewhere in these forums but I'm trying to avoid that.  Since the O3AS work is GPU-only I can imagine some uncertainty, at the server, whether to send opencl or cuda to a particular host that is capable of both.  Maybe the opencl (1.07) version is considered the default/preferred app; but that brings me back to my first question:  How do I prod the servers to send me the 1.14 (cuda) app?

I have set NTT and in 24 hours the cache will be empty.  Then "reset project" and see if that results in both cuda and opencl apps being downloaded.  Fingers crossed...

 

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6441
Credit: 9571527114
RAC: 8342070

I can post my current working

I can post my current working app_info.xml

I have one that I was running with an rtx 3080 ti that will allow All-Sky GW, brp7/meerKat and a CPU app grp#5.

Tom M

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4964
Credit: 18720785871
RAC: 6411638

The O3ASHF1b_0.config

The O3ASHF1b_0.config and O3ASHF1b_1.config files can be just created empty with an editor.  But they need to be defined and called out in the app_info.xml file AS WELL as entries in the client_state.xml file for the <app_version> sections in the client_state file for the app.

 

Eugene Stemple
Eugene Stemple
Joined: 9 Feb 11
Posts: 67
Credit: 374671491
RAC: 543234

>Tom M I can post my

>Tom M

I can post my current working app_info.xml

   ... hold that thought ...

After the "reset project" early this morning, the servers sent me (cpu) apps for FBRP5 and BRP4G and a cuda (gpu) app for BRP7 but only the opencl (gpu) app v1.07 for O3AS.  The two O3AS..*.. config files were downloaded also.  But -so far- no cuda app v1.14 for O3AS.  I'll give it a few more days but I'm not optimistic that the server will ever send v1.14 . 

That leaves the app_info.xml method as the only solution.  I have hesitated to do that because it makes the app updates a manual process rather than the server initiated automatic process.  (Somebody will correct me if that is a misunderstanding on my part...) 

I think there may be an app_info.xml in another thread but Tom M posted a pre-1.14 file in #221367 of this thread (Jan 18) which should be a good enough guide for me (with follow-up discussion regarding the *2.config file being extraneous).   But one step at a time... let's give the servers a little more time to send v1.14 automatically.  As mentioned in a previous post, I do have the v1.14 executable already downloaded manually, ready to reference in the app_info.xml when/if the time comes.

 

Eugene Stemple
Eugene Stemple
Joined: 9 Feb 11
Posts: 67
Credit: 374671491
RAC: 543234

Good progress...   I let

Good progress...   I let things run for 2 days, hoping the servers would notice my system is cuda capable and would download O3AS 1.14, but no luck with that strategy.  So, I built an app_info.xml for just the one O3AS 1.14 app and after a project reset and two minor bumps (emoticon:slap forehead) it is downloading and running the cuda app as desired and results are being validated.   I can see a path forward to add other apps to the app_info.xml and get back to running all the Einstein apps as I was before.  But I have to wonder if I can drop the app_info.xml, now that the pump has been primed, so to speak, with a valid client_state.xml and a "known" O3AS cuda app in place, and see if the servers will be happy to keep sending work units for the 1.14 cuda app.  It's a long shot, I guess.  But no way to know for sure without trying it.  So that's the plan... draining the cache today and after saving the whole boinc directory I'll restart boinc with the app_info.xml removed.   (I might note that the Berkeley on-line help regarding the app_info.xml file format was not much help - browsing these forum posts for examples was more useful to me.)

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6441
Credit: 9571527114
RAC: 8342070

My own experience is it will

My own experience is it will default to 1.08

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Jimbocous
Jimbocous
Joined: 31 Mar 17
Posts: 67
Credit: 1222422625
RAC: 1700053

Tom M wrote: My own

Tom M wrote:

My own experience is it will default to 1.08

That's what I saw, as well. Posted a message about that here quite a while ago but never saw any response.

 

 

 

 

Eugene Stemple
Eugene Stemple
Joined: 9 Feb 11
Posts: 67
Credit: 374671491
RAC: 543234

This was a dead end.  The

This was a dead end.  The servers downloaded the opencl 1.07 app and resent a bunch of "lost tasks", which are crunching away as OpenCL work.  There was a line in the event log to "see scheduler log messages" and in the host log I am unsure what to make of it.  There is a line ...checking plan class GW-opencl-nvidia-2 ... followed a couple of lines later with ...plan class o.k. ... then "Don't need CUDA jobs, skipping version 1.07"   After that there is a line ...checking plan class GW-cuda ... followed by  ... plan class o.k. ... then "Don't need CUDA jobs, skipping version 1.15 ..."  and ... no app version available APP#56 PLATFORM#7 ...

Sure seems like version 1.07 is the default.  Am I missing something in cc_config or coproc parameters that makes it look like I don't want cuda work?  When I enable the MeerKat/BRP7 app I do get the cuda app and work units downloaded automatically.

I've run out of ideas.  Any more thoughts?  Or just proceed with the app_info.xml method for anonymous platforms to run the 1.14 cuda app.

 

Comment viewing options

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