Mantis Bug Tracker
Main | My View | View Issues | Change Log | Roadmap
  

View Issue Details [ Jump to Notes ] [ Issue History ] [ Print ]
IDProjectCategoryView StatusDate SubmittedLast Update
0000956LuxRenderCorepublic2011-03-02 08:072012-09-11 06:11
ReporterJ the Ninja 
Assigned ToOmniflux 
PrioritynormalSeverityminorReproducibilityalways
StatusclosedResolutionfixed 
PlatformOSOS Version
Product Version 
Target Version1.0Fixed in Version1.0RC4 
Summary0000956: Resuming from an FLM makes a mess of "total" stats
DescriptionThe FLM apparently does not include how long it was rendered for. Upon resuming, the stats system appears to believe all of the samples in the FLM were computed during those few seconds of preprocessing. If the render was run for any significant amount of time, the total samples per second stat becomes utterly useless, as it will report values several multiples of what is actually happening.

If you add a network node, things get even worse. It appears to attribute the samples from the resumed film as being sent back by the network node since the render was resumed, meaning the network samples per second will also be far higher than they actually are. Furthermore, this causes Lux to incorrectly estimate the rate at which samples are being taken, since it is overestimating the network contribution. As a result, the total samples per pixel value is only correct right after samples have been collected from network nodes.

Possible solution:
1. Include time rendered in the FLM, and sum these values when merging films that are not from a network node, so a correct total running time is available to stats system (this would probably require adding some sort of "this one was sent back from network node" flag in the film as well)

2. Fix whatever is causing Lux to believe the samples from the film came from a network node when estimating the sampling rate for the render farm.
TagsNo tags attached.
Mercurial Changeset #b70a5a4b2ccd
Requires Documentation UpdateNo
Requires Exporter Update
Attached Files

- Relationships

-  Notes
(0002858)
Lord Crc (administrator)
2011-03-29 04:20

 In order to not make a complete mess of backwards compatibility, I suggest we introduce two new film parameters:
- A float to hold the number of seconds elapsed on this rendering, using the secElapsed renderer statistic.
- A bool to indicate if this flm is intended for network transmission and not a "regular" resume file.

Introducing these as parameters means you can load them up in older version without lux barfing (just an error message), and it also makes it easy to override these values which likely will be useful.
(0002860)
jeanphi (administrator)
2011-03-29 05:52

 I think we should first find a sane semantic of using the value, and I don't think we're going to succeed.
I rather think that each node should compute a real time S/s value and the interface should display the sum.
Samples loaded at the beginning should not be part of the computation since they can come from an entirely different setup and adjusted stats might be completely meaningless.

- Issue History
Date Modified Username Field Change
2011-03-02 08:07 J the Ninja New Issue
2011-03-18 01:58 jeanphi Target Version => 0.8RC2
2011-03-19 05:59 jeanphi Target Version 0.8RC2 => 0.8RC3
2011-03-29 04:14 Lord Crc Assigned To => Lord Crc
2011-03-29 04:14 Lord Crc Status new => assigned
2011-03-29 04:20 Lord Crc Note Added: 0002858
2011-03-29 05:52 jeanphi Note Added: 0002860
2011-04-17 11:11 jeanphi Target Version 0.8RC3 => 0.8
2011-06-06 08:21 jeanphi Target Version 0.8 => 1.0
2012-04-11 18:43 Omniflux Mercurial Changeset # => b70a5a4b2ccd
2012-04-11 18:43 Omniflux Assigned To Lord Crc => Omniflux
2012-04-11 18:43 Omniflux Status assigned => resolved
2012-04-11 18:43 Omniflux Resolution open => fixed
2012-04-11 18:43 Omniflux Fixed in Version => 1.0
2012-08-21 06:54 jeanphi Fixed in Version 1.0 => 1.0RC4
2012-09-11 06:11 jeanphi Status resolved => closed

Copyright © 2000 - 2012 MantisBT Group
Powered by Mantis Bugtracker