Page 1 of 1

Network speed question

Posted: Tue Aug 02, 2016 7:55 pm
by mttvngln
Long time lurker, first time poster. First of all, thank you for making this program! It has been a real life saver for me. I've been using it for a couple years now and have gotten excellent results with it. This forum has helped me with a lot of the growing pains in the beginning.

My question concerns my new setup of a bunch of old editing machines that I've converted into a transcode farm. I have 3 of them as nodes, all networked together through a 5 port home unmanaged gigabit switch (pretty low quality, I admit). Due to the network set up we have there is a managed fibre switch running the editing system and a decent unmanaged gigabit switch connected to a few other systems. I have a NAS straddling both networks so both groups (designers and editors) can see the same shared storage. FFAStrans is not running off that shared storage, only off the 3 nodes with a shared striped pair of drives in one. I have another program, DropIt, grabbing files from the NAS and moving them to the transcode main node and then grabbing the finished files and moving them back to the NAS.

The issue is that when I had each node set to process up to 5 files at once the processing speed/fps of each file was nearly nothing, maybe 2 fps. When it was running on a single computer we could get about 3 jobs going at once at more or less equivalent fps to the file, so for our purposes around 30fps. After I had the system doing a few files at once we were doing okay, we lost a bit of individual transcode speed but were able to transcode 3-4 files at once instead of waiting. Once I put it in a stress situation ~20 files, XDCAM 50 mxf transcoding to a broadcast MPEG-TS (custom ffmpeg) all nodes dropped to that terrible processing speed I mentioned, with almost no CPU use. About 20% of the files didn't fail in transcode but produced an AVS file instead of the expected mpeg.
This is, other than the new set up, a tried and true workflow.

Once I changed each node from 5 jobs to 1 there was a significant increase in speed across the board, but still not what it was before, around 20fps.

Is this because of my not great network set up? I know each individual machine can handle what I was throwing at it. The shared drive is a striped pair of drives, lots of speed. I don't have the budget for transceivers to partake of the very nice fibre switch, but perhaps I can cut that low quality switch out of the equation. I had thought that any information the 3 nodes need to share would be shared within their little environment and not out into the larger system due to the location of the main node, but I must be wrong about that.

Sorry for the wall of text, any suggestions you have would be great.

Re: Network speed question

Posted: Wed Aug 03, 2016 2:59 pm
by admin
Hi mttvngln, nice of you to finally join in then :-)

Can you please share some more details on how you have set up your farming environment? It's not clear to me.

Low frame rate and low cpu usage will normally indicate slow reading/writing of files. And this is likely to be the case the more files being processed simultaneously.

Failing to deliver the proper file is obviously a bug and I've fixed a lot of them in the upcoming 0.7.6 version.

-steinar

Re: Network speed question

Posted: Thu Aug 04, 2016 2:01 pm
by mttvngln
Thanks!

This is the set up:
PC A (shared drive, FFAStrans node), PC B (FFAStrans node), PC C (FFAStrans node) --> 5 port gigabit switch --> gigabit switch (designers) --> NAS (mirror of shared folder for the editors)

If you're saying it's read/write maybe I'm not getting as much speed as I thought on those striped drives.

Re: Network speed question

Posted: Thu Aug 04, 2016 9:54 pm
by admin
Well, I think that makes sense. The more processes trying to access a drive, the slower each read/write access will be. And the fact that you get higher fps the fewer simultaneous jobs you run proves (in my opinion) that this is the case. Now, I'm not saying it IS a shared drive issue, it might also be a network bottleneck. Maybe you can find some test tools to check your infrastructure?

But if I get you setup correctly; the work root has been set to shared (PC A) on all 3 hosts, right?

-steinar