-
FCPX NFS tuning
Hi everyone!
First: I’ve been a reader of this forum since quite a while and would like to thank everyone for this nice resource of insight!
Now I have a question you guys might be able to help:
The current setup I administrate (in Education…) is using a MacMini as fileserver with 2 Promise Pegasus attached (via thunderbolt, obviously), 1GbE and multiple MacPros as editing stations. Original media is stored centrally on the fileserver via AFP. Everything is running on latest 10.9. We’re mainly editing FullHD ProRes 422 with Multicam content using up to 6 cameras.
Originally we used the sparsebundle-workflow with the sparsebundles hosted on the fileserver to make libraries available on all editing stations, but after a while, this turned out to be insufficient. Editing a multicam-project becomes slow as soon as FCPX has to render some content (due to any effect or whatever) and doesn’t access the original media anymore. So we switched over to place the libraries locally on the editing stations.
I only learned recently that NFS is actually supported to host libraries in a network storage, so right now I’m trying to get it running. It basically works, but also doesn’t perform well. As long as I’m playing back original media (multiclip but no effects) which are accessed over afp it plays back smoothly, but as soon as FCPX renders content (due to color correction or whatever) and has to access those files from the library (this time over NFS) there are lags of 1-2 seconds and frame dropouts of sometimes >50%.
The Promises still have lots of “headroom”, the fileserver doesn’t max on cpu or ram, the ethernet network should also handle the traffic (copying files both over NFS and AFP are very close to 120MBytes/s, also activity monitor never maxes the ethernet connection while editing).
Do you have any tips on how to improve performance in this setup? I really don’t see what’s the bottleneck in this setup and it’s getting kind of frustrating ;-).
We plan to upgrade the uplink fileserver -> switch to 10GbE, but with a single client in the testing environment this shouldn’t be necessary yet, right?
Thank you for any tips!
Benni Graf.