New Improved Gravational Wave App - Discussion

mikey
mikey
Joined: 22 Jan 05
Posts: 12679
Credit: 1839081536
RAC: 3902

Eugene Stemple wrote: Or

Eugene Stemple wrote:

Or just proceed with the app_info.xml method for anonymous platforms to run the 1.14 cuda app.

That's what I would do, easy and quick and it will do exactly what you want, just keep a note so you remember WHY it's doing that when you want to change things.

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6435
Credit: 9564342478
RAC: 9042908

mikey wrote: Eugene Stemple

mikey wrote:

Eugene Stemple wrote:

Or just proceed with the app_info.xml method for anonymous platforms to run the 1.14 cuda app.

That's what I would do, easy and quick and it will do exactly what you want, just keep a note so you remember WHY it's doing that when you want to change things.

+1

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!

walton748
walton748
Joined: 1 Mar 10
Posts: 94
Credit: 1485664350
RAC: 2162860

Another way might be do

Another way might be to deinstall the opencl-icd through apt and restart your system. boinc will not detect any opencl capacity, thus not report it to the servers, with cuda still intact. I think after that you should be able to switch between cuda and cuda-beta via host preferences, as you wish to. You will loose the ability to use opencl for any other project you wish to run in parallel, though.

I cannot acces my computers until in a few days to prove my point, unfortunately.

 

Cheers,

Walton

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6435
Credit: 9564342478
RAC: 9042908

walton748 wrote: Another way

walton748 wrote:

Another way might be to deinstall the opencl-icd through apt and restart your system. boinc will not detect any opencl capacity, thus not report it to the servers, with cuda still intact. I think after that you should be able to switch between cuda and cuda-beta via host preferences, as you wish to. You will loose the ability to use opencl for any other project you wish to run in parallel, though.

I cannot acces my computers until in a few days to prove my point, unfortunately.

 

Cheers,

Walton

Walton,

That is a GREAT idea!.

I wish I was flexible enough to think of that.

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: 4963
Credit: 18709501314
RAC: 6332514

This would be the fastest and

This would be the fastest and easiest solution.  After a deinstall and reboot, problem solved.  But you would need to run your cache of 1.07 tasks down or you will abort all of them on the reboot and then be put in the penalty box for too many errors.

 

Eugene Stemple
Eugene Stemple
Joined: 9 Feb 11
Posts: 67
Credit: 373802464
RAC: 544803

walton748 wrote: Another way

walton748 wrote:

Another way might be to deinstall the opencl-icd through apt and restart your system. boinc will not detect any opencl capacity, thus not report it to the servers, with cuda still intact. I think after that you should be able to switch between cuda and cuda-beta via host preferences, as you wish to. You will loose the ability to use opencl for any other project you wish to run in parallel, though.

I found a similar/equivalent method...

In  cc_config.xml, set this parameter...

  <no_opencl>1</no_opencl>

And, also, in the app_config.xml, in the project directory, include an <app_version> section that includes:

<app_name>einstein_O3AS</app_name>

<plan_class>GW-cuda</plan_class>

   Restarting the boinc client after those edits the O3AS v1.07 (opencl) app is no longer used and work for it no longer downloaded.  Upon examining the server log I see a message to the effect "... only test applications are available for this plan_class..." so I went to the host application preferences and re-enabled "run test applications".  I don't have an <app_versions> section for the BRP7 app so it is happy to run its cuda version (and the other -asteroids- project also has a cuda app) so the blocking of opencl does not adversely affect my boinc usage.  And, if need be, there is a workaround by running a second instance of boinc+boinc-manager with its own host name and cc_config settings.  So far, however, I'm getting BRP7 cuda work and FGRP5 cpu work but nothing for O3AS.  I'll give the servers more time... if I'm asking for "test app" O3AS work maybe that is not issued very often; or if the beta test cycle has ended maybe I'll never get "test app" work :(

Is anybody without test apps enabled getting O3AS cuda work?  Either v1.08 or v1.14 or v1.15 ?

The sched_request still shows the v1.07 app although new work is (apparently) blocked by the <no_opencl> config setting.  And the client_state[prev].xml files also show the v1.07 app files.  Nothing anywhere, of course, for v1.14 since the server never sent me that app.

 

 

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4963
Credit: 18709501314
RAC: 6332514

There is no 1.08 app anymore

There is no 1.08 app anymore in the apps availability.  That got changed to beta 1.15.  There still is 1.14 app available without beta being enabled.

I'm still getting 1.14 O3AS work with no issues via an app_info.

2024-06-25 07:17:27.5080 [PID=645516]    Anonymous platform app versions:
2024-06-25 07:17:27.5080 [PID=645516]       app: einsteinbinary_BRP7 ver: 999
2024-06-25 07:17:27.5080 [PID=645516]       app: einstein_O3AS ver: 114
2024-06-25 07:17:27.5199 [PID=645516]    [mixed] sending locality work first (0.2015)
2024-06-25 07:17:27.5236 [PID=645516]    [send] send_old_work() no feasible result older than 336.0 hours
2024-06-25 07:17:27.9270 [PID=645516]    [version] [HOST#12291110] anonymous app 'einstein_O3AS' version num 114 on platform 'x86_64-pc-linux-gnu'
2024-06-25 07:17:27.9270 [PID=645516]    [version] Found anonymous platform app for einstein_O3AS: plan class GW-cuda
2024-06-25 07:17:27.9285 [PID=645516]    [send] [HOST#12291110] [WU#813465988 h1_1531.60_O3aC01Cl1In0__O3ASHF1d_1532.00Hz_45959] using delay bound 604800 (opt: 604800 pess: 604800)
2024-06-25 07:17:27.9290 [PID=645516] [debug]   Sorted list of URLs follows [host timezone: UTC-25200]
2024-06-25 07:17:27.9290 [PID=645516] [debug]   zone=-21600 url=http://einstein-dl.phys.uwm.edu
2024-06-25 07:17:27.9290 [PID=645516] [debug]   zone=-18900 url=http://einstein-dl.syr.edu
2024-06-25 07:17:27.9290 [PID=645516] [debug]   zone=+03600 url=http://einstein2.aei.uni-hannover.de/EinsteinAtHome
2024-06-25 07:17:27.9473 [PID=645516]    [send] est. duration for WU 813465988: unscaled 51038.35 scaled 5170.27
2024-06-25 07:17:27.9474 [PID=645516]    [HOST#12291110] Sending [RESULT#1626521052 h1_1531.60_O3aC01Cl1In0__O3ASHF1d_1532.00Hz_45959_1] (est. dur. 5170.27 seconds, delay 604800, deadline 1719904647)
2024-06-25 07:17:29.9592 [PID=645516]    [send] est. duration for WU 813466033: unscaled 51038.35 scaled 5170.27
2024-06-25 07:17:29.9592 [PID=645516]    [send] [WU#813466033] meets deadline: 5170.27 + 5170.27 < 604800
2024-06-25 07:17:29.9593 [PID=645516]    [send] [HOST#12291110] [WU#813466033 h1_1531.60_O3aC01Cl1In0__O3ASHF1d_1532.00Hz_45958] using delay bound 604800 (opt: 604800 pess: 604800)
2024-06-25 07:17:29.9975 [PID=645516]    [send] est. duration for WU 813466033: unscaled 51038.35 scaled 5170.27
2024-06-25 07:17:29.9976 [PID=645516]    [HOST#12291110] Sending [RESULT#1626521175 h1_1531.60_O3aC01Cl1In0__O3ASHF1d_1532.00Hz_45958_0] (est. dur. 5170.27 seconds, delay 604800, deadline 1719904649)

I'd just reset your host back to stock to clear up your client_state.xml file section for the project.  You should get the stock 1.14 app and tasks then with the no_opencl setting in the cc_config.

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6435
Credit: 9564342478
RAC: 9042908

Maybe this will be

Maybe this will be helpful.

<app_version>
    <app_name>einstein_O3AS</app_name>
    <version_num>114</version_num>
    <platform>x86_64-pc-linux-gnu</platform>
    <avg_ncpus>1.000000</avg_ncpus>
    <plan_class>GW-cuda</plan_class>
    <api_version>7.3.0</api_version>
    <file_ref>
        <file_name>einstein_O3AS_1.14_x86_64-pc-linux-gnu__GW-cuda</file_name>
        <main_program/>
    </file_ref>
    <file_ref>
        <file_name>O3ASHF1b_0.config</file_name>
    </file_ref>
    <file_ref>
        <file_name>O3ASHF1b_1.config</file_name>
    </file_ref>
    <coproc>
        <type>NVIDIA</type>
        <count>1.0</count>
    </coproc>
    <dont_throttle/>
  </app_version>

-------------

Maybe adding   <version_num>114</version_num> someplace?

I think you are still trying to avoid running with an app_info.xml file?

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!

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3945
Credit: 46701142642
RAC: 64300495

in the app_info file, you

in the app_info file, you don't even need the version number line at all. omit it completely and it's easier to switch back and forth between app versions just by changing the filename called.

if you leave the app version in, and later change which app you want to use without exhausting the cache, BOINC will delete all of them when you make the switch since all the old tasks in the queue are marked for the 'old' app but BOINC doesnt see (app_info) that you have the 'old' app anyway and abandons them. without the parameter at all, they all get marked with v0.00 regardless of which one is actually in use and BOINC wont abandon them.

_________________________________________________________________________

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4963
Credit: 18709501314
RAC: 6332514

Neat solution with one flaw. 

Neat solution with one flaw.  If you are using BoincTasks standard or JS version, BT doesn't know how to handle tasks enumerated with 0.00 version numbers.

Some end up at the front of sorting, some end up with phantom numbering from other task types from the project or other strangeness.

I posted on the BT forum about the issue but he can't recreate the problem because he never implemented the actual app_info I assume.

 

Comment viewing options

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