- Wed May 03, 2006 4:09 pm
#149373
Well I was hoping to post some results from my last attempt at cooperative rendering but I've had some problems. I was rendering with 80+ nodes and have discovered some things that make using cooperative rendering with this many machines very difficult.
-The auto merge feature will only work if all the render nodes that started the job are online and working properly. Apparently cooperative rendering is unable to bypass nodes that have had problems or have gone off line.
-It's not possible to remove or add a render node to a job once it's been started. This makes job management very difficult.
-When you create render farm groups, those groups will only exist as long as the manager is running; once it is shut down the groups disappear.
-Some of my render nodes still show that they are rendering long after the target time has expired, this holds up the entire merger process until the last machine is finished. There should be a way to bypass those machines that are either still rendering or have errored out or gone down so that you can still get a merged MXI file.
-It isn’t possible to pause a job
-Once one job is finished you have to clear out all MXI and MXS files from the folder and re-boot the node in order to render a new scene. If you don’t do this the node will only render the fist scene it received, no new scenes can be rendered.
-There is no useful data generated by mxcl –d or the mxcl –manager as far as error reports so it’s almost impossible to know why a particular job failed on any node.
-There is no way to preview the cooperative rendering job since the image file being saved to the network directory seems to only come from a single render node and the merger process only occurs after the entire job has completed rendering.
-Node folders are left with hundreds of Megs worth of useless data in them that must be manually cleaned up after every rendering.
Over all my first large scale cooperative rendering test has been a complete failure. There is no way that anyone can effectively use cooperative rendering with a rendering farm of any size. There must be some kind of job management built into the system so that problems can be dealt with manually, but there also needs to be some automatic management. I would like to know how the A-team and NL tested this system and how they set it up in order to get around these problems because from where I stand there is no way this can be considered a usable process.
-The auto merge feature will only work if all the render nodes that started the job are online and working properly. Apparently cooperative rendering is unable to bypass nodes that have had problems or have gone off line.
-It's not possible to remove or add a render node to a job once it's been started. This makes job management very difficult.
-When you create render farm groups, those groups will only exist as long as the manager is running; once it is shut down the groups disappear.
-Some of my render nodes still show that they are rendering long after the target time has expired, this holds up the entire merger process until the last machine is finished. There should be a way to bypass those machines that are either still rendering or have errored out or gone down so that you can still get a merged MXI file.
-It isn’t possible to pause a job
-Once one job is finished you have to clear out all MXI and MXS files from the folder and re-boot the node in order to render a new scene. If you don’t do this the node will only render the fist scene it received, no new scenes can be rendered.
-There is no useful data generated by mxcl –d or the mxcl –manager as far as error reports so it’s almost impossible to know why a particular job failed on any node.
-There is no way to preview the cooperative rendering job since the image file being saved to the network directory seems to only come from a single render node and the merger process only occurs after the entire job has completed rendering.
-Node folders are left with hundreds of Megs worth of useless data in them that must be manually cleaned up after every rendering.
Over all my first large scale cooperative rendering test has been a complete failure. There is no way that anyone can effectively use cooperative rendering with a rendering farm of any size. There must be some kind of job management built into the system so that problems can be dealt with manually, but there also needs to be some automatic management. I would like to know how the A-team and NL tested this system and how they set it up in order to get around these problems because from where I stand there is no way this can be considered a usable process.
fantastic the glowing teapot...Dunno looks like coop needs some 'fixin, I'd suggest heavy drinking/tranquilizers while you manually merge the files right now. Looks like its gots potential, though
Keep emitters as low poly as possible. If you can, keep them a single plane, but in this case i think a 9 poly sphere will be enough.
good news, I've determined that you don't have to erase the cooperative.mxi files from the node directories. As long as you restart mxcl -server you will be fine and can render out a new scene.
- By Mark Bell
- By Matteo Villa