Manual 'branding' for stock Seti apps ?


log in

Advanced search

Message boards : Number crunching : Manual 'branding' for stock Seti apps ?

1 · 2 · Next
Author Message
Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 967744 - Posted: 3 Feb 2010, 1:20:57 UTC

How can I change a WU from 609 (cuda23) to 603?

I have looked at the ReScheduler from Lunatics but I am not sure that it will work with the stock Seti apps.

I tried stopping Boinc and then making the changes to the <workunit> ... <app_version>nnn field but, the WU's went back to the GPU when I rebooted and restarted Boinc.

Could anyone point me to more info about what is required to re-brand a WU when using the stock apps?

Profile ignorance is no excuse
Avatar
Send message
Joined: 4 Oct 00
Posts: 9529
Credit: 44,433,321
RAC: 0
Korea, North
Message 967754 - Posted: 3 Feb 2010, 2:35:59 UTC - in response to Message 967744.

since the rescheduler is an 3rd party app. and it is the creation of the wonderful folks ar Lunatics, then you might head to their website., register, and ask the question there. they created it they will know best
____________
In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

End terrorism by building a school

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 967768 - Posted: 3 Feb 2010, 3:52:50 UTC - in response to Message 967754.

since the rescheduler is an 3rd party app. and it is the creation of the wonderful folks ar Lunatics, then you might head to their website., register, and ask the question there. they created it they will know best

Thanks, I'm sure they will know best ... I'm registered there but I suspect they will suggest that I just dl and run the optimized apps ... then their Resched will work better.

Just curious to know if anyone here might know the 'simple' steps to do it manually for people like me who still use the stock apps.

Profile hiamps
Volunteer tester
Avatar
Send message
Joined: 23 May 99
Posts: 4292
Credit: 72,971,319
RAC: 0
United States
Message 967770 - Posted: 3 Feb 2010, 4:01:16 UTC - in response to Message 967768.

since the rescheduler is an 3rd party app. and it is the creation of the wonderful folks ar Lunatics, then you might head to their website., register, and ask the question there. they created it they will know best

Thanks, I'm sure they will know best ... I'm registered there but I suspect they will suggest that I just dl and run the optimized apps ... then their Resched will work better.

Just curious to know if anyone here might know the 'simple' steps to do it manually for people like me who still use the stock apps.


I don't believe you can...Thats why the Lunatics do it....
____________
Official Abuser of Boinc Buttons...
And no good credit hound!

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 967776 - Posted: 3 Feb 2010, 4:45:15 UTC - in response to Message 967770.

I don't believe you can...Thats why the Lunatics do it....


You might be *very* right about that! We shall see ...

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 967781 - Posted: 3 Feb 2010, 5:37:05 UTC - in response to Message 967776.

You might be *very* right about that! We shall see ...


It's beginning to look like Re-scheduling tasks between the CPU and GPU is only going to work with the 'anonymous platform' mechanism.

Like you said, it doesn't look like I can get there from here with the stock apps.

Thanks. Looks like I'll just have to let my cache run out ... and then checkout the optimized goodies to put an app_info file in place.

Josef W. SegurProject donor
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4332
Credit: 1,113,463
RAC: 995
United States
Message 967793 - Posted: 3 Feb 2010, 7:21:32 UTC - in response to Message 967781.

You might be *very* right about that! We shall see ...


It's beginning to look like Re-scheduling tasks between the CPU and GPU is only going to work with the 'anonymous platform' mechanism.

Like you said, it doesn't look like I can get there from here with the stock apps.

Thanks. Looks like I'll just have to let my cache run out ... and then checkout the optimized goodies to put an app_info file in place.

An app_info.xml is not needed, only BOINC's client_state.xml is examined and modified. What you have apparently missed is that GPU tasks have <plan_class>cuda</plan_class> and CPU tasks have no plan_class.

If stock apps are what you want to run, Raistmer's Perl script ought to work and probably Marius' executable too. But I think both are set to reschedule between 6.03 CPU and 6.08 GPU (rather than 6.09 GPU).
Joe

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 967800 - Posted: 3 Feb 2010, 8:06:38 UTC - in response to Message 967793.

An app_info.xml is not needed, only BOINC's client_state.xml is examined and modified. What you have apparently missed is that GPU tasks have <plan_class>cuda</plan_class> and CPU tasks have no plan_class.

If stock apps are what you want to run, Raistmer's Perl script ought to work and probably Marius' executable too. But I think both are set to reschedule between 6.03 CPU and 6.08 GPU (rather than 6.09 GPU).
Joe


Thanks Joe. I knew I missing something in the translation ...

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8757
Credit: 52,707,223
RAC: 27,580
United Kingdom
Message 967808 - Posted: 3 Feb 2010, 9:42:18 UTC

You need to change three things in order to make the rebranding work:

* Change version number in <workunit>
* Change version number in <result>
* Remove <plan_class> from <result>

The tricky bit is matching up the <workunit> and the <result>. You need to do them in matching pairs.

Marius, who wrote the rebranding tool, hasn't posted since September 2009, when he wrote "... I'm swamped in work the last months. There's very little time left for seti at this moment.": so if you want to rebrand stock 609 applications, you'll probably have to adapt one of the scripts yourself.

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 967809 - Posted: 3 Feb 2010, 9:45:36 UTC - in response to Message 967793.
Last modified: 3 Feb 2010, 9:49:19 UTC

If stock apps are what you want to run, Raistmer's Perl script ought to work and probably Marius' executable too. But I think both are set to reschedule between 6.03 CPU and 6.08 GPU (rather than 6.09 GPU).


I see the <plan_class>cuda23</plan_class> in the <results> for 609 but, not in the <workunit> section.

I'll just let my cache run out and then install the optimized apps. I suspect it will be a little safer that way.

I had 'given up' thinking about it because I noticed that my tasks are assigned either for 603 or 609 (cuda23) when I list them. Seems like it might be important to make sure I return the result as '609 (cuda23)' if that's the way it had been assigned. I was concerned that it might mess up the database otherwise. That's why I'd figured it might not be possible without using the anonymous platform.

I noticed that with the anonymous platform the tasks don't seem to be assigned to a specific <plan_class> ...

I'll just go with the optimized apps within a few days.

Thanks again everyone. I'm just trying to figure out how things work under the hood.

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 967810 - Posted: 3 Feb 2010, 9:51:56 UTC - in response to Message 967808.
Last modified: 3 Feb 2010, 9:53:04 UTC

You need to change three things in order to make the rebranding work:

* Change version number in <workunit>
* Change version number in <result>
* Remove <plan_class> from <result>

The tricky bit is matching up the <workunit> and the <result>. You need to do them in matching pairs.

Thanks Richard ... I'll dig into it a little deeper after I've switched to the optimized apps in a few days.

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8757
Credit: 52,707,223
RAC: 27,580
United Kingdom
Message 967811 - Posted: 3 Feb 2010, 10:24:30 UTC - in response to Message 967809.

I noticed that with the anonymous platform the tasks don't seem to be assigned to a specific <plan_class> ...

You'll see exactly the same assignment in BOINC Manager (and the underlying client_state.xml) with the optimised apps - it's only the website listing which changes.

But we've been returning tasks issued to CUDA, crunched by CPU for months now, and I don't think that's anything to do with SETI's database problems ;-)

Profile BilBg
Volunteer tester
Avatar
Send message
Joined: 27 May 07
Posts: 2879
Credit: 6,450,117
RAC: 2,950
Bulgaria
Message 968042 - Posted: 4 Feb 2010, 8:09:30 UTC


"Easy" direct link to ReSchedule 1.9

http://atsplus.ru/boinc/ReSchedule1.9.rar


____________



- ALF - "Find out what you don't do well ..... then don't do it!" :)

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 968045 - Posted: 4 Feb 2010, 8:29:23 UTC - in response to Message 968042.

Thanks, I'll give it a shot after I get the optimized apps ... I think I had noticed this at Lunatics too.

Cheers

Profile Odan
Send message
Joined: 8 May 03
Posts: 91
Credit: 15,331,177
RAC: 0
United Kingdom
Message 968074 - Posted: 4 Feb 2010, 14:10:59 UTC - in response to Message 967811.

I don't remember this being discussed but it probably has:

Couldn't SETI just not issue VLAR/VHAR WUs to CUDA & keep them for CPU? I don't know enough about the details of how the back end works but rather ramble on for a couple of pages I'll leave it at that.

Heaven forbid that anyone think I am suggesting anything time consuming for the already over-busy guys at Berkley to do! It would get over one of the reasons to re-brand but not the BOINC confusion with 2 different apps running the same WUs on the same machine leading to a dearth of GPU WUs.

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8757
Credit: 52,707,223
RAC: 27,580
United Kingdom
Message 968101 - Posted: 4 Feb 2010, 16:40:44 UTC - in response to Message 968074.

I don't remember this being discussed but it probably has:

Couldn't SETI just not issue VLAR/VHAR WUs to CUDA & keep them for CPU? I don't know enough about the details of how the back end works but rather ramble on for a couple of pages I'll leave it at that.

Heaven forbid that anyone think I am suggesting anything time consuming for the already over-busy guys at Berkley to do! It would get over one of the reasons to re-brand but not the BOINC confusion with 2 different apps running the same WUs on the same machine leading to a dearth of GPU WUs.

Wow, that's almost an anniversary present. Prompted me to look up the text of an email I sent to Eric Korpela on 6 February 2009.

The main problem with CUDA at the moment seems to be the user experience of
screen 'stutter' and lagging while the GPU is processing VLAR tasks. The
extended runtime is unfortunate, of course, but I doubt many users would
notice it on its own: it's the display problem which draws users to the
message boards to express a feeling of dis-satisfaction with CUDA.

It seems as if the final resolution of the whole VLAR issue is taking longer
than anticipated. We had a brainstorming session at Lunatics this morning,
including most members of this distribution list: if there had been any
further news from Jorden's NB (copied into this email by bcc), I think it
would have been mentioned.

So, I make the following proposal to alleviate volunteer dis-satisfaction in
the short term, and buy additional breathing space for the long term
development.

Stop sending VLAR tasks to CUDA clients.

That sounds drastic, but I believe it would be surprisingly easy to do. The
key fact is this:

All true VLAR tasks have
<rsc_fpops_est>80360000000000.000000</rsc_fpops_est>

The scheduler has access to <rsc_fpops_est> (used to estimate task duration)
and could use it as a filter - something like:

CASE <rsc_fpops_est> == 80360000000000.000000
SET app_max_version = 603

- in other words, ensure that VLAR gets sent to a traditional (CPU)
application.

Advantages:
1) Greatly increased user satisfaction with CUDA
2) Slight reduction in result error rate (fewer aborts/autokills)
3) Slightly reduced bandwith usage (fewer resends)
4) Raistmer can stop playing with autokill mods, and go back to
optimising (!)

Disadvantages:
1) Code complexity, fiddling with scheduler - inevitable potential for
mistakes
2) Lower scheduler efficiency, possible impact on feeder queue.
3) <add your own here>

I am least able to quantify disadvantage (2), but I don't imagine it would
be too great. There should still be plenty of demand from stock 603
applications to soak up the likely number of VLARs, so the feeders shouldn't
clog up: and - as I was explaining to Matt in technical news last night -
the Arecibo 'weather forecast' for the next month suggests that task and
bandwidth demand will be lower than recently, because the 'basketweave'
survey has been replaced by a 'drift' survery thru March 12 (last date in
current public observing schedule).

One obvious point - the proposed scheduler change should be applied to the
Main project only: Beta stills needs the full range of ARs sent to CUDA apps
for testing.

Richard Haselgrove

I never got a reply.

Profile Odan
Send message
Joined: 8 May 03
Posts: 91
Credit: 15,331,177
RAC: 0
United Kingdom
Message 968108 - Posted: 4 Feb 2010, 17:42:01 UTC - in response to Message 968101.
Last modified: 4 Feb 2010, 17:45:59 UTC

Richard, I could always edit my post out & try again on Saturday if that would make you happy :)

It does sound as easy as I hoped it would be. Never mind, eh?

I did like your comment about stopping Raistmer and playing with autokill!

ClaggyProject donor
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 4207
Credit: 34,463,683
RAC: 20,317
United Kingdom
Message 968171 - Posted: 5 Feb 2010, 0:42:35 UTC - in response to Message 968101.

Strangely enough this Changeset appeared today: Changeset 20419

- scheduler: add example code for SETI@home's situation


where app has both GPU and CPU versions,
but for certain jobs (VLAR WUs in this case)
the GPU version performs poorly and shouldn't be used.


The fix is a kludge - it will result in these jobs
not being sent to the host at all,
rather than being sent with the CPU app.
The current architecture makes it difficult to do otherwise.
One possible fix would be to create a separate app
for VLAR jobs, with only CPU app versions.

Claggy

Profile [seti.international] Dirk Sadowski
Volunteer tester
Avatar
Send message
Joined: 6 Apr 07
Posts: 7115
Credit: 61,255,496
RAC: 7,221
Germany
Message 968172 - Posted: 5 Feb 2010, 0:57:17 UTC - in response to Message 967744.
Last modified: 5 Feb 2010, 0:58:26 UTC

How can I change a WU from 609 (cuda23) to 603?

I have looked at the ReScheduler from Lunatics but I am not sure that it will work with the stock Seti apps.

I tried stopping Boinc and then making the changes to the <workunit> ... <app_version>nnn field but, the WU's went back to the GPU when I rebooted and restarted Boinc.

Could anyone point me to more info about what is required to re-brand a WU when using the stock apps?



Just curious..

Why you don't want to use the opt. project applications?

You would double the performance of your CPU.
Also you would increase the performance of your GPU (with VLARkill).

There is also a noVLARkill mod.
[http://lunatics.kwsn.net/index.php?module=Downloads;catd=9]_CudaMB V12 non-VlarKill (EXE Only, Advanced Users)
IIRC, it should be the same version like in the Lunatics Installer (but without VLARkill).
Both have a ~ 50 % faster CUDA WU preparation time on the CPU.

You could use the opt. app for CPU and the noVLARkill app for GPU. If you don't like to 'kill' CUDA WUs.
And then maybe the ReScheduler..


____________
[Optimized project applications, for to increase your PC performance (double RAC)!][Overview of abbreviations, which are used often in forum and their meaning.]
____________
BR

SETI@home Needs your Help ... $10 & U get a Star!

Team seti.international

Das Deutsche Cafe. The German Cafe.

Profile Phil J Taylor
Avatar
Send message
Joined: 26 Dec 09
Posts: 96
Credit: 603,521
RAC: 0
United States
Message 968176 - Posted: 5 Feb 2010, 1:31:56 UTC - in response to Message 968172.
Last modified: 5 Feb 2010, 1:41:35 UTC

Just curious..

Why you don't want to use the opt. project applications?

You would double the performance of your CPU.
Also you would increase the performance of your GPU (with VLARkill).


I just installed the optimized apps and changed to default 'VlarKill' app to the 'noKill' variety. I'm also running Reschedule manually so, problem solved.

I had 'backed my truck up to the loading dock' on Saturday this week when I had noticed a lot of 'shorties' being sent out. Silly me, I had asked for about 4-days worth of work to cover me with the stock apps through the outage on Tuesday. I did get a lot of the quick work units but it also loaded me up with about 2-weeks worth of Vlars for my CUDAs ... so, I was trying to find a way to move them to the CPU with the stock apps. No cigars ...

Instead, I just manually 'aborted' all the extra Vlars, let my cache run down and, then installed the optimized apps so that I could use Reschedule.

I doubt that my overall performance will double with the optimized apps because I am not using 'VlarKill' ... I do suspect that my RAC will get a 30-40% boost over time though. I've wanted to use the stock apps while I get to know my way around a bit. My i7 950 is nearing it's peak around 7500 with the stock apps. I'll be very happy if I see it rise above 10k with the optimized apps and my two little GTS 220's ;)

1 · 2 · Next

Message boards : Number crunching : Manual 'branding' for stock Seti apps ?

Copyright © 2014 University of California