Sunflower A.12.12.2.L (slow)

Sunflower A.12.12.2.L (slow)

Description

Destruction below
Frank is flying but suddenly notices all the motion going on below as the traps unfold themselves.
This session is particularly slow.

This is the left eye view of the stereo video. The right eye is rendered in Session 1473.

Big Buck Bunny is copyright 2008, Blender Foundation / www.bigbuckbunny.org available under CC-by v3

2054836

Message boards : Comments and discussion : 1476

Author Message
Loup
Send message
Joined: 18 Feb 12
Posts: 17
Credit: 3,016,533
RAC: 0
Message 11354 - Posted: 22 May 2012, 21:47:59 UTC
Last modified: 22 May 2012, 21:49:03 UTC

There seem to be a few issues here. Look at http://burp.renderfarming.net/workunit.php?wuid=1695410:
It again waited for a third result to then pick two identical ones without trying to get a third "good" one => shouldn't it wait for three valid ones? It is also funny that it didn't report any result mismatch before the third result arrived (it usually reported something like "done but unclear result" or something, don't remember it clearly).

The credit system also tried to be funny again and decided that ~4 days of rendering are worth the same then your typical 30 minute render job - either something in its calculations overflowed or it must have a really really complex basis to get to that conclusion ... ;)

Apart from that I would also be interested in what was wrong with the result of http://burp.renderfarming.net/result.php?resultid=7077778, that PC usually doesn't cause any problems. Or is the validator again acting up because of random "noise" in the output?

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4507
Credit: 2,094,806
RAC: 0
Message 11355 - Posted: 23 May 2012, 15:49:16 UTC

It does indeed sound a bit interesting.

I'll wait with the analysis of the results until the rest of the instances are done. The two first frames were sent out to specifically test how the linux hosts would react to them and to see what kind of noise profile the results would have.

Tore Zachariassen
Send message
Joined: 30 Mar 07
Posts: 1
Credit: 29,415
RAC: 0
Message 11406 - Posted: 30 Jun 2012, 20:49:57 UTC

After more than 180h of computing (on all 4 CPUs)- and closing in on this long WU (over 80%), a power failure forced me to restart the computer... and of course - BURP restarted at 0,00 %.
Sorry - but I have to say this - this is NOT acceptable - and I will abort the WU within the next few minutes...
Please - sort the checkpoint-issue out... before releasing this extremely long WUs...

tuankiet65
Send message
Joined: 7 Jun 12
Posts: 4
Credit: 76,612
RAC: 0
Message 11462 - Posted: 6 Sep 2012, 10:26:11 UTC - in response to Message 11406.

OMG. Bad thing for you
____________

noderaser
Project donor
Avatar
Send message
Joined: 28 Mar 06
Posts: 514
Credit: 1,564,255
RAC: 0
Message 11463 - Posted: 11 Sep 2012, 4:22:21 UTC

Unfortunately, checkpointing is something that the Blender folks have to take care of.
____________
Click here to see My Detailed BOINC Stats

Loup
Send message
Joined: 18 Feb 12
Posts: 17
Credit: 3,016,533
RAC: 0
Message 11470 - Posted: 29 Sep 2012, 17:55:55 UTC - in response to Message 11463.

Apart from taking a long time per job which will be an issue for some people, validation really seems to be an issue here again:
http://burp.renderfarming.net/workunit.php?wuid=1698611

It is one thing to get >4 results for a short task but for these units the whole thing will never complete this way (not to mention that needlessly rendering stuff for days is not exactly ideal). Espescially since it is a rarity that just three results are returned for a result, let alone 5 or more.

I will keep doing these but I wouldn't be surprised to see the whole session still running in 2014 the way it is progressing right now. :(

Speedy
Send message
Joined: 25 May 06
Posts: 206
Credit: 674,159
RAC: 0
Message 11471 - Posted: 29 Sep 2012, 21:11:45 UTC - in response to Message 11470.

Has any managed to render a eorkunit using a VM under windows? If yes what was the configuration used?
____________

Have a crunching good day!!
Live in NZ y not join Smile City?

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4507
Credit: 2,094,806
RAC: 0
Message 11479 - Posted: 2 Oct 2012, 17:17:53 UTC - in response to Message 11470.

Apart from taking a long time per job which will be an issue for some people, validation really seems to be an issue here again:
http://burp.renderfarming.net/workunit.php?wuid=1698611

Yeah it caught my eye as well. I'll have to pull out some of the returned results to do a comparison to figure out whether this is simply tolerable noise (in which case the validator can be temporarily "softened") or whether it is actual pixel errors. Crossing my fingers for the former - hopefully tomorrow we will know.

@Speedy:
Yup, I did try a 64-bit Ubuntu install in a VirtualBox once. With a 50GB dynamically sized disk it seemed to do quite well even with Sunflower tasks. It isn't perfect, but at least it is a sneaky way of getting checkpointing.

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4507
Credit: 2,094,806
RAC: 0
Message 11484 - Posted: 4 Oct 2012, 19:46:43 UTC

Luckily it looks like the differences in results are fairly minor - but enough to confuse the validator. I turned off the automatic validation for any new workunits created for this session. I'll update the already running workunits with the new values before the weekend.

Speedy
Send message
Joined: 25 May 06
Posts: 206
Credit: 674,159
RAC: 0
Message 11485 - Posted: 4 Oct 2012, 21:04:30 UTC - in response to Message 11479.

Thanks Janus, How much ram did you allow in the VM/ That's interesting you could process a Sunflower unit at the same time. Did the sunflour unit take longer than usually to complete while you were running one in VM at the same time?
____________

Have a crunching good day!!
Live in NZ y not join Smile City?

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4507
Credit: 2,094,806
RAC: 0
Message 11486 - Posted: 6 Oct 2012, 10:11:14 UTC - in response to Message 11485.

I think it was around 9.5GB and the Sunflower unit was running inside the VM not outside - didn't notice much of a difference in speed but I assume it was slower. I did, however, notice that the host OS became slower at doing I/O stuff (like copying files).

Speedy
Send message
Joined: 25 May 06
Posts: 206
Credit: 674,159
RAC: 0
Message 11490 - Posted: 6 Oct 2012, 21:35:43 UTC - in response to Message 11486.

I think it was around 9.5GB and the Sunflower unit was running inside the VM not outside - didn't notice much of a difference in speed but I assume it was slower. I did, however, notice that the host OS became slower at doing I/O stuff (like copying files).

OK Thanks 9.5GB for a VM on my pc may be a bit tight. I have 12GB total using 7 Ultimate for the OS
____________

Have a crunching good day!!
Live in NZ y not join Smile City?

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4507
Credit: 2,094,806
RAC: 0
Message 11497 - Posted: 11 Oct 2012, 16:00:10 UTC
Last modified: 11 Oct 2012, 16:00:46 UTC

For now the validator will have marked one result as valid for each frame from this session even though more than one result was returned successfully. This will arbitrarily assign credit to that one result rather than the other ones. The remaining credit will be granted on "Pay Day". Additionally the silly credit system seemed to grant pretty much random amounts of credit when it did actually grant something - this too will be corrected at a later time.
Good news is that the session is moving forward again.

Speedy
Send message
Joined: 25 May 06
Posts: 206
Credit: 674,159
RAC: 0
Message 11498 - Posted: 12 Oct 2012, 5:26:30 UTC - in response to Message 11479.

@Speedy:
Yup, I did try a 64-bit Ubuntu install in a VirtualBox once.

When you tried it what VM software did you (eg Oracle VM VirtualBox) use & did you install ubuntu?
____________

Have a crunching good day!!
Live in NZ y not join Smile City?

Speedy
Send message
Joined: 25 May 06
Posts: 206
Credit: 674,159
RAC: 0
Message 11499 - Posted: 20 Oct 2012, 0:23:24 UTC

I'm up & running with Ubuntu 12.10 in Oracle VM VirtualBox 4.1.22 r80657. Thanks for all the advice :)
____________

Have a crunching good day!!
Live in NZ y not join Smile City?

JZD
Send message
Joined: 30 Dec 11
Posts: 95
Credit: 3,446,583
RAC: 10
Message 11506 - Posted: 12 Nov 2012, 20:34:20 UTC

Can someone explain how is it possible that this workunit 1715067 count 85 hodin and get it for 58.74 kredits, workunit 1776303 count 72 hodin a tyhle workunit 1776299 count 53 hodin and get it for 6,876.42 kredits and workunit 1776290 count 55 hodin and get it for 9,962.15 kredits. I can see some discrepancy. How so?

AMDave
Send message
Joined: 22 Apr 05
Posts: 4
Credit: 102,536
RAC: 0
Message 11580 - Posted: 23 Jan 2013, 14:57:44 UTC - in response to Message 11506.
Last modified: 23 Jan 2013, 15:00:16 UTC

Error while computing
Run time: 534,148.74 seconds
CPU time: 2,656,494.63 seconds
SunflowerBlender (Linux) v4.84 (mt)

VERY Unhappy.
Once upon a time this project did not suck.
____________

noderaser
Project donor
Avatar
Send message
Joined: 28 Mar 06
Posts: 514
Credit: 1,564,255
RAC: 0
Message 11581 - Posted: 24 Jan 2013, 4:31:44 UTC

That seems a rather extreme response to one failed WU...
____________
Click here to see My Detailed BOINC Stats

Speedy
Send message
Joined: 25 May 06
Posts: 206
Credit: 674,159
RAC: 0
Message 11582 - Posted: 24 Jan 2013, 8:04:48 UTC
Last modified: 24 Jan 2013, 8:10:44 UTC

I have to agree with AMDave 148.374 hours (elapsed time) 6.182 days is a heck of a lot of time to have a task fail
____________

Have a crunching good day!!
Live in NZ y not join Smile City?

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4507
Credit: 2,094,806
RAC: 0
Message 11591 - Posted: 27 Jan 2013, 13:41:22 UTC - in response to Message 11580.
Last modified: 27 Jan 2013, 13:41:53 UTC

Sorry about the workunit - especially since it was almost done.
The workunit didn't fail, though. BOINC told it to abort and it did:

Exception caught: BOINC forcibly tells us to abort right now!


Exactly what caused BOINC to ask BURP to abort this unit is a good question - typically stuff like other processes using all the memory, other processes causing freezes so that BOINC thinks BURP is dead etc.

At least we are now down to the last handful of these extremely long sessions.


Post to thread

Message boards : Comments and discussion : 1476