log in

Advanced search

Message boards : Number crunching : RESOLVED - 2 tasks both consume CPU time

1 · 2 · 3 · Next
Author Message
Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 91 - Posted: 28 Jan 2008, 22:55:42 UTC
Last modified: 28 Jan 2008, 23:01:35 UTC

Is MindModeling supposed to be a dual task application or is the second task a visualisation task that should rather sit still when not in use?

Currently I have 3 running tasks on a dual CPU machine, one is SIMAP, the other two are sbcl.exe and ACTR_2.61_windows_intelx86.exe, both belonging to MindModeling.

Each of the 3 tasks receive about 1/3rd of the total CPU time (average) of both CPUs.


p.s.: this might cause the strange behaviour while the project is paused as well as strange credit claims, if it doesn\'t add the CPU time of both tasks. BOINC cannot handle such a thing correctly yet.

p.p.s.: Yep, it does only count the CPU time consumed by one of the two tasks, sbcl.exe, whereas the ACTR CPU time is ignored.

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 92 - Posted: 28 Jan 2008, 23:03:48 UTC

I started a thread here http://boinc.berkeley.edu/dev/forum_thread.php?id=2449, maybe they can include such a project type into their concepts somehow.

Profile Cori
Avatar
Send message
Joined: 25 Jan 08
Posts: 29
Credit: 920
RAC: 0
Message 93 - Posted: 28 Jan 2008, 23:07:39 UTC
Last modified: 28 Jan 2008, 23:07:59 UTC

Btw: there\'s a third app called watchdog.exe which is also running with MindModelling. What does this one do?
And when I noticed the errors on my Win NT host the watchdog.exe was still running when the WUs had already errored out, strange...
____________
Greetings from Cori

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 95 - Posted: 28 Jan 2008, 23:11:42 UTC
Last modified: 28 Jan 2008, 23:14:13 UTC

Watchdog consumes no CPU time.

watchdog will ensure that a process does not continue if the parent process ends.

Usage: watchdog ParentPID PotentialOrphanedPID


Funny that especially this process survives a crashed result *g

Parent process ID must be the ACTR...exe, sbcl.exe the one to watch for.

Profile Cori
Avatar
Send message
Joined: 25 Jan 08
Posts: 29
Credit: 920
RAC: 0
Message 96 - Posted: 28 Jan 2008, 23:15:33 UTC - in response to Message 95.
Last modified: 28 Jan 2008, 23:16:07 UTC

... Funny that especially this process survives a crashed result *g

Hehe... and it also was resistant to resetting the project. Had to delete it manually from the project\'s folder. ;-)
____________
Greetings from Cori

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 97 - Posted: 28 Jan 2008, 23:17:13 UTC

Useless watchdogs risk to end up on a yummy Satay picker ;-)

Profile Cori
Avatar
Send message
Joined: 25 Jan 08
Posts: 29
Credit: 920
RAC: 0
Message 100 - Posted: 28 Jan 2008, 23:21:00 UTC - in response to Message 97.
Last modified: 28 Jan 2008, 23:21:32 UTC

Useless watchdogs risk to end up on a yummy Satay picker ;-)


*ROFL* Or as a so-called hot dog... :-)



*yum yum*
____________
Greetings from Cori

Profile Rebirther
Avatar
Send message
Joined: 25 Jan 08
Posts: 17
Credit: 4,439
RAC: 0
Message 101 - Posted: 28 Jan 2008, 23:32:38 UTC - in response to Message 95.

Watchdog consumes no CPU time.

watchdog will ensure that a process does not continue if the parent process ends.

Usage: watchdog ParentPID PotentialOrphanedPID


Funny that especially this process survives a crashed result *g

Parent process ID must be the ACTR...exe, sbcl.exe the one to watch for.


watchdog comes from rosetta ;)

Profile Jack.Harris
Project administrator
Project developer
Project scientist
Avatar
Send message
Joined: 24 Apr 07
Posts: 507
Credit: 761,261
RAC: 0
Message 102 - Posted: 28 Jan 2008, 23:40:04 UTC - in response to Message 101.

Watchdog consumes no CPU time.

watchdog will ensure that a process does not continue if the parent process ends.

Usage: watchdog ParentPID PotentialOrphanedPID


Funny that especially this process survives a crashed result *g

Parent process ID must be the ACTR...exe, sbcl.exe the one to watch for.


watchdog comes from rosetta ;)


I don\'t know anything about the rosetta watchdog.
I wrote the MM@Home watchdog to ensure sbcl does not hang around if the ACTR BOINC wrapper application is killed [or if God forbids...dies]
--Jack

Profile Cori
Avatar
Send message
Joined: 25 Jan 08
Posts: 29
Credit: 920
RAC: 0
Message 103 - Posted: 28 Jan 2008, 23:41:27 UTC - in response to Message 101.
Last modified: 28 Jan 2008, 23:43:54 UTC

Watchdog consumes no CPU time.

watchdog will ensure that a process does not continue if the parent process ends.

Usage: watchdog ParentPID PotentialOrphanedPID


Funny that especially this process survives a crashed result *g

Parent process ID must be the ACTR...exe, sbcl.exe the one to watch for.


watchdog comes from rosetta ;)


They can keep it. :-P
Btw: when I first saw the name in my task manager I feared it was malware, because I usually don\'t like watchdogs on my puters. *LOL*


EDIT: Whoops, didn\'t see your post. Ok, I don\'t kill the poor dog anymore. ;-)
____________
Greetings from Cori

Profile Jack.Harris
Project administrator
Project developer
Project scientist
Avatar
Send message
Joined: 24 Apr 07
Posts: 507
Credit: 761,261
RAC: 0
Message 104 - Posted: 28 Jan 2008, 23:43:24 UTC - in response to Message 103.

Watchdog consumes no CPU time.

watchdog will ensure that a process does not continue if the parent process ends.

Usage: watchdog ParentPID PotentialOrphanedPID


Funny that especially this process survives a crashed result *g

Parent process ID must be the ACTR...exe, sbcl.exe the one to watch for.


watchdog comes from rosetta ;)


They can keep it. :-P
Btw: when I first saw the name in my task manager I feared it was malware, because I usually don\'t like watchdogs on my puters. *LOL*


Maybe I should call it mm_watchdog.exe
--Jack

Profile Cori
Avatar
Send message
Joined: 25 Jan 08
Posts: 29
Credit: 920
RAC: 0
Message 105 - Posted: 28 Jan 2008, 23:44:25 UTC - in response to Message 104.

They can keep it. :-P
Btw: when I first saw the name in my task manager I feared it was malware, because I usually don\'t like watchdogs on my puters. *LOL*


Maybe I should call it mm_watchdog.exe
--Jack

That is a good idea. Thanks! :-)
____________
Greetings from Cori

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 108 - Posted: 29 Jan 2008, 0:06:35 UTC
Last modified: 29 Jan 2008, 0:26:42 UTC

Now it started 2 MM workunits, i.e. 4 tasks consuming CPU time.

sbcl.exe gets less than 10%, ACTR a little more than 40% ... * 2 as it is a 2 CPU machine

So if sbcl.exe is the actual working task, it is really inefficient (10% - 20%) because ACTR steals most of the CPU time - and BOINC counts only the little CPU time that is used up by sbcl.exe :-/


If ACTR is just waiting for the end of sbcl.exe, it should not poll but sleep while waiting for a signal that the child process died.

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 110 - Posted: 29 Jan 2008, 0:44:42 UTC
Last modified: 29 Jan 2008, 0:53:56 UTC

update : The two ACTR are now using nearly all CPU time, both sbcl.exe gets some 1%-2%, \"% done\" is nearly stuck.

I already gave both sbcl.exe a process priority above normal, they still do not consume CPU time.


After a little more than 50 minutes, the results are showing 14% progress and 7.5 minutes used CPU time each.

The task manager shows : both ACTR have used 45 minutes, both sbcl.exe 7.5 minutes.

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 112 - Posted: 29 Jan 2008, 1:01:43 UTC
Last modified: 29 Jan 2008, 1:08:29 UTC

one more update : the system(!) load is at 80%, user load only at about 20%

There is really something wrong because usually it should be at 1% - 2% (maximum) whereas the user load should be greater than 98%

It is not HD activity (swapping or so) which would cause the system load to go up too. It must be something different, a loop with permanent kernel calls or so.

I will reset the project for now, this leads nowhere.

I saved all files from one of the \"slots\" directories, if you need them for debugging, I can zip them and put them on the web for download.

Profile Jack.Harris
Project administrator
Project developer
Project scientist
Avatar
Send message
Joined: 24 Apr 07
Posts: 507
Credit: 761,261
RAC: 0
Message 113 - Posted: 29 Jan 2008, 1:40:50 UTC - in response to Message 112.

one more update : the system(!) load is at 80%, user load only at about 20%

There is really something wrong because usually it should be at 1% - 2% (maximum) whereas the user load should be greater than 98%

It is not HD activity (swapping or so) which would cause the system load to go up too. It must be something different, a loop with permanent kernel calls or so.

I will reset the project for now, this leads nowhere.

I saved all files from one of the \"slots\" directories, if you need them for debugging, I can zip them and put them on the web for download.


If sbcl isn\'t taking nearly 100% then the process is paused. This can occur explicitly through the BOINC Manager or implicitly through preference settings [user interaction].
The way it works is that the ACTR_BOINC.exe will unsuspend sbcl when appropriate. It polls the shared memory for communication provided to it by the BOINC Mananger [between polling the process sleeps].

With no user interaction the sbcl process will monopolize things. On windows you will the ACTR_graphics jump in every now and then (less if proper OpenGL drivers are installed) and the ACT_BOINC process will monitor to see if it should suspend things etc.

Hope this helps,
--Jack

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 124 - Posted: 29 Jan 2008, 7:50:55 UTC
Last modified: 29 Jan 2008, 8:18:41 UTC

No graphics (I don\'t use screensaver or BOINC graphics window at all), no pause settings (BOINC can use the PC anytime unthrottled), no restrictions in using the ressources, no competing programs eating CPU seconds.

You can see it here as well :

http://mindmodeling.org/beta/forum_thread.php?id=21

Look at the screenshot (not my screenshot), there you can see what ACTR does and what sbcl.exe does.

While one MM task was running together with one SIMAP, it looked the same on my box (see how this thread started).

When there were two MM tasks with no other project active, both ACTR jumped to even much higher CPU load and it got worse after half an hour until sbcl.exe used nearly no CPU time anymore.


p.s.: SIMAP is a BOINC project with very low requirements, it doesn\'t need much RAM and it doesn\'t stress the memory bus.


p.p.s.: Not to be misunderstood, MM did use 100% CPU but most power went into ACTR, not sbcl.exe. The system load (red bars in the CPU usage graph on windows) was extremely high at > 80%, the user load (green bars in the CPU usage graph, which should mostly represent the crunching itself) was less than 20%

Profile Ananas
Send message
Joined: 24 Jan 08
Posts: 33
Credit: 100,084
RAC: 0
Message 140 - Posted: 29 Jan 2008, 18:59:51 UTC
Last modified: 29 Jan 2008, 19:01:58 UTC

Just in case you want to have a look into the slot contents ...

http://oct31.de/tmp/MM_2process_problem.zip

(init_data.xml anonymized)

If there is a switch to get more debug output somehow, I would like try that too, if it helps.
I could do a standalone (BOINCless) run as well, if you tell me how to do that.

Dagorath
Send message
Joined: 29 Jan 08
Posts: 29
Credit: 4,047
RAC: 0
Message 148 - Posted: 29 Jan 2008, 21:08:07 UTC

The same thing happened here on my single core, Win XP machine... Task Manager shows ACTR consumes about 49% of CPU, sbcl consumes about 49%. The result where I observed that is here, my computer is 864. The task details page claims only 1 hour CPU time was used whereas my ../boinc/stdoutae.txt shows the task ran for 2 hours on the wall clock. Since there were no other CPU intensive processes running while that task crunched, the wall clock time and CPU time should be very close but they are not.

I noticed the same on one of my Linuxes boxes... top reports ACTR using 49% of CPU, sbcl using 49%.

Profile Jack.Harris
Project administrator
Project developer
Project scientist
Avatar
Send message
Joined: 24 Apr 07
Posts: 507
Credit: 761,261
RAC: 0
Message 153 - Posted: 29 Jan 2008, 21:39:29 UTC - in response to Message 148.

The same thing happened here on my single core, Win XP machine... Task Manager shows ACTR consumes about 49% of CPU, sbcl consumes about 49%. The result where I observed that is here, my computer is 864. The task details page claims only 1 hour CPU time was used whereas my ../boinc/stdoutae.txt shows the task ran for 2 hours on the wall clock. Since there were no other CPU intensive processes running while that task crunched, the wall clock time and CPU time should be very close but they are not.

I noticed the same on one of my Linuxes boxes... top reports ACTR using 49% of CPU, sbcl using 49%.



What is your BOINC client version?
--Jack

1 · 2 · 3 · Next

Message boards : Number crunching : RESOLVED - 2 tasks both consume CPU time


Main page · Your account · Message boards


Copyright © 2020 MindModeling.org