You need to open the Compressor App and open the Help menu to read the user manual, especially the Distributed Processing Setup section of the manual (click on the text on the first page of the user manual). There is a Qmaster 2 manual:
https://manuals.info.apple.com/en/Apple_Qmaster_2_User_Manual.pdf, but most of the information in it is more relevant to Shake and Maya clusters, and the Distributed Processing Setup manual now contains most (all?) of its contents and more stuff relevant to Qmaster use with Compressor.
> But do each of the other computers need to have access to the video clips?
Yes, since they have to have access to transcode the video. You can get the job-submitting client to copy all the source video to the Cluster Storage, so freeing up the job-submitting client’s bandwidth. If your job-submitting client is also the Cluster Controller and Cluster Storage, then all computers will need access to that computer.
>If so, does this only work when using an Xsan or fiber network of stored media,
> where all the networked computers are also able to look at the clips?
Well, it can work with any network, as long as “The network must support the Apple networking technology built in to Mac OS X.” (from the Compressor user manual). The manual mentions the possible use of 100-Base-T networking.
The Cluster Storage is a central location which can contain the original video, temporary intermediate files, and the final output of the distributed storage. To do all three, the access needs to be set to Read & Write for each node in the cluster. You can also set the location for temporary files to be a different location, and this would be especially important when dealing with a slow file networking system.
Page 35 of the Compressor 2 user manual describes how you can set whether the source video get copied to the Cluster Storage. Note carefully page 44, 112 and 113 where it describes the limitations with segmented jobs (i.e. multiple computers handling the same piece of video, instead of multiple computers handling successive encoding jobs). This may have an effect on whether it is worthwhile rendering on a cluster if you are using mulltiple pass VBR.
Bill Lee