Simple Processing Over 1gb Network Unbearably Slow

[ratio]

I work in a mostly flame-based facility, but we use shake for a lot of utility - image conversion, temp comps, etc. (as well as some final comps).

The whole network (OS X, Linux, SGI) is on Gigabit Ethernet network.

When converting 2k cineons to 1k proxy jpegs in a shake script, when fileIn images are on the network, the processing takes an average of 35 seconds/frame (in the command line). However, when you pull the cineons locally (by drag-and-drop in the OS X Finder), the file transfer is extremely quick. Then, actually rendering the file conversion locally takes more like .6 seconds/frame.

So if the network is theoretically fast, and the CPUs are theoretically fast, why the extremely slow render time over the network?

[deke-kincaid]

try using the -nocache option so it isn't looking for the file locallally when rendering. You are command line rendering right?

[ratio]

QUOTE(beaker @ 11/27/07, 11:36 AM) [snapback]277079[/snapback]
try using the -nocache option so it isn't looking for the file locallally when rendering. You are command line rendering right?

Upon your suggestion, I tried using -nocache in the exec. (Yes, I'm rendering from the command line - thanks for checking.)

Unfortunately, render time didn't improve at all. I asked our sysadmin just now and he said that there's something funny about how shake handles i/o stuff with NFS file system. I, not being a sysadmin, have no idea what that means.

I asked sysadmin if there was a way to fix it. He said no. He said that the server with the source files was really loaded up and constantly hit hard and he needed to do some cleanup.

Does anyone out there know any better? Is there some more efficient way for shake to read and write files on NFS?

[deke-kincaid]

maybe try mounting the drive with smb?