Net render: speed, optimization

doctorfish wrote on 10/26/2004, 2:00 AM
I've been reading some of the older posts on net rendering and was wondering what kind of luck people were having with it lately.

I recently set up a network (a 10 mb/s connection through a highspeed router) to test the net rendering and had the following results.

Net rendered to windows media video took 10 1/2 hours including stitch time.

The same project on one machine took only 7 1/2 hours.????????

Has anyone else experienced thus?

What optimizations and tweaks do you do to get better results with net rendering?

Dave

Comments

jester700 wrote on 10/26/2004, 4:09 AM
A Gigabit network would help... ;-)
Chienworks wrote on 10/26/2004, 4:21 AM
Consider that DV is about 37.5Mb/s and uncompressed video/audio data comes in at around 248.5Mb/s. On a 10Mb/s connection it can take 4 or 5 minutes to copy 1 minute of DV or over 30 minutes to copy 1 minute of uncompressed. 10Mb/s just isn't fast enough for video work. The networked computers end up sitting around twidding their thumbs waiting for the next package of bits to process.
GaryKleiner wrote on 10/26/2004, 9:19 AM
I have had some very good results with long renders.

One thing you can do, if you have hyperthreaded machines, is to set up two render nodes on each machine.

I go into this in detail on The Vegas 5 Companion DVDs.

Gary
johnmeyer wrote on 10/26/2004, 11:51 AM
You need 100 Mbps. 10 Mbps won't cut it at all. Also, the project you are rendering must required some really heavy-duty rendering (bezier masks, parent/child composting, motion blur, supersampling, etc.). If it just has transitions and a little color correction, the reduction in rendering time will be killed by all the file transfers over the network, and especially by the long stitching times.
BrianStanding wrote on 10/26/2004, 1:36 PM
John,

My second computer finally died, so I can no longer test this myself.

Have you tried using a networked machine as a render engine WITHOUT checking the "distributed render" checkbox? I can imagine some circumstances where it might make sense to set up a render queue on a "render-only" machine while continuing to edit or make revisions on my primary machine. Or if you have a project with discrete segments where you want two separate rendered files for your final output, you could set up each machine to render its own segment.

This would seem to avoid the "stitching" slowdown at the end. Or does it?

Just curious.
skibumm101 wrote on 10/26/2004, 1:52 PM
Related question. I have a hyperthreading processor, Would i get additional benefits by installing a node on the local machine. i donot have any machines that are render nodes,
doctorfish wrote on 11/12/2004, 5:19 PM
Thanks to all who replied and my apologies for being so slow to respond to my own post. I have done any more tests recently but I thinking that it might be the network that was just too slow. Brian Standing has the same idea that I developed after my initial tests. I'm planning to use the remote machine for rendering while I continue to edit on my main machine.

That way at least I can continue to work while rendering is going on.

Dave
BJ_M wrote on 11/12/2004, 6:24 PM
there is some merit at times to cancel a net render (remembering to keep all the pre-stitched files) when finished, but before stitching and stitch them yourself to another dirve .. on big big projects i have found this to work quicker.. or to read/write to a raid 0 fast as possable array ..
johnmeyer wrote on 11/13/2004, 8:22 AM
ere is some merit at times to cancel a net render (remembering to keep all the pre-stitched files) when finished, but before stitching and stitch them yourself to another dirve

Can this be done reliably? Do you just click on "cancel" as soon as it reports that stitching is being done? If it has been stitching for five minutes before I click on cancel, are some of the files lost?

If you can tell me how to do this, it would really make a big difference. If I can get at the unstitched files, then I should be able to bring them directly into Vegas and never stitch them at all, but instead simply print to tape, or render from the timeline. Someone suggested this last spring, here in the forum, and it was a most brilliant suggestion. Once you think about it, given that Vegas can print from the timeline, and render from the timeline, why the heck should we have to stitch? Further, why doesn't Sony just build a VEG file during the network render that automatically loads all the files on the timeline?
BrianStanding wrote on 11/18/2004, 12:41 PM
Can't you just arbitrarily cut the timeline in half, render each half on a separate networked machine ("render loop region only" and UNCHECK "distribute render"), then manually create a new Veggie file with both halves butted up against each other?

Not as automatic as it should be, perhaps, but you still get the benefit of two machines working simultaneously: one machine working on the first half, the other machine working on the second half, and no stitching at the end.

I haven't benchmarked it, but it seems considerably faster on my network with an XP 2700 desktop and a P4 2.8 laptop than it is with just the XP 2700 alone.