ValidateME

ValidateME

Description

This short fly-by anim was used to validate the validator
To test out the new validator-code this short animation was produced using more or less random editting methods, inserting random lightsources at random times etc.
Nothing useful can possibly come out of this, but it\'ll probably look interesting and burn some CPU cycles along the way to test each part of the renderpipe and the validator at high resolution.

Message boards : Comments and discussion : 340

Author Message
axarydax
Send message
Joined: 8 Mar 07
Posts: 11
Credit: 1,604
RAC: 0
Message 5041 - Posted: 23 Mar 2007, 13:38:18 UTC

This looks like a fat workunit :)

So when this passes, the validator would be ok?

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4483
Credit: 2,094,806
RAC: 0
Message 5057 - Posted: 23 Mar 2007, 18:18:19 UTC

So far the validator has passed the positive test (ie. if two results are identical it validates them as identical).
When the session is done we will see if it passes the negative test (ie. if two results differ it should not validate them as identical).

Profile Wabbit98
Send message
Joined: 5 Mar 05
Posts: 8
Credit: 2,646
RAC: 0
Message 5073 - Posted: 24 Mar 2007, 6:25:32 UTC

So after this will there be more work sessions in quick succession or long period of downtime between sessions?
____________

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4483
Credit: 2,094,806
RAC: 0
Message 5076 - Posted: 24 Mar 2007, 9:24:47 UTC - in response to Message 5073.

So after this will there be more work sessions in quick succession or long period of downtime between sessions?

After this we\'ll resume \"normal alpha operation\" - meaning that sessions are rendered if there are any in the queue.

Profile AndyK
Project donor
Avatar
Send message
Joined: 2 Apr 05
Posts: 137
Credit: 20,063
RAC: 0
Message 5084 - Posted: 24 Mar 2007, 14:44:30 UTC - in response to Message 5057.

So far the validator has passed the positive test (ie. if two results are identical it validates them as identical).
When the session is done we will see if it passes the negative test (ie. if two results differ it should not validate them as identical).


how do you want to cause a session to report different results for a frame/part?
And will it be casual differences or systematic?

Andy
____________

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4483
Credit: 2,094,806
RAC: 0
Message 5089 - Posted: 24 Mar 2007, 16:24:21 UTC - in response to Message 5084.
Last modified: 24 Mar 2007, 16:30:58 UTC

Mistakes happen from time to time. For the past sessions a few parts have been returned with invalid content (probably due to overclocking on the machines that returned them) or missing content whatsoever.
Since the part of the server that receives files from BOINC has always checked for these issues, a frame that contained such a faulty part could not be automatically completed (and a cleanup script was run to select the properly rendered part later instead).

So since there\'s now a proper validator in place there should no longer be any of these orphaned workunits and everything should be completely automatic.

Also a look into the log-files for the validators (there\'s 2 of them) should show delta values deviating from zero for the parts with issues.

If there are workunits that the validator validated but the file archiver doesn\'t like then the validator has failed the test. The only situation I could imagine that this would happen is if I\'ve made a coding error in the validator - it is ment to use the same strict checking-scheme as the file archiver.

Profile Izarf
Send message
Joined: 29 Jul 05
Posts: 33
Credit: 13,637
RAC: 0
Message 5135 - Posted: 28 Mar 2007, 7:17:41 UTC

Nice job, Janus! =D
____________

axarydax
Send message
Joined: 8 Mar 07
Posts: 11
Credit: 1,604
RAC: 0
Message 5142 - Posted: 28 Mar 2007, 13:30:04 UTC

goood.

Fortunately blender is deterministic, isn\'t it? I can\'t imagine situation when one frame would render differently on two computers - could it happen when we outrule the hardware errors (overclocking, etc.)?

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4483
Credit: 2,094,806
RAC: 0
Message 5148 - Posted: 28 Mar 2007, 18:16:23 UTC - in response to Message 5142.

Fortunately blender is deterministic, isn\'t it? I can\'t imagine situation when one frame would render differently on two computers - could it happen when we outrule the hardware errors (overclocking, etc.)?

It has happened before but only due to bugs in Blender - and they are usually fixed pretty fast when reported (the Blender devs are pretty active people).
Of course it is possible to introduce real randomness to a session by using scripted stuff, but people who know how to do this also know why it\'s a bad idea...

Profile Janus
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 16 Jun 04
Posts: 4483
Credit: 2,094,806
RAC: 0
Message 5171 - Posted: 30 Mar 2007, 16:02:23 UTC

This session concludes the validator test, part of the results are publiced here.

The graphical output of this session is pretty odd but it achieved the point of triggering some of the faulty machines out there to return faulty results.


Post to thread

Message boards : Comments and discussion : 340