[Sidefx-houdini-list] Mantra farm rendering to server killing network, ideas?

Matt Estela matt.estela at gmail.com
Wed Sep 10 02:09:49 EDT 2014

In case anyone is interested, we had a go at implementing this via the
mantra equivalent of an rifilter, seems to be doing the job rather well.
details here:


On 9 September 2014 00:26, Antoine Durr <antoinedurr at gmail.com> wrote:

> It shouldn't be too hard to implement this in the wrapper script that
> renders the Mantra jobs.  Assuming you've written out the .ifd for the
> frame, just find and replace the output file with a local filename, then
> render away, then copy the frame wholesale.  You'll want to 'touch' the
> original file location just to be sure that it can be properly written to.
> The disadvantage of this approach is that you can't preview frames in
> progress as they're all on their local render hosts.  But the network
> savings might be worth it, though of course you need to do the basic math:
> 200 machines * 120 MB / seconds-per-frame = X MB/sec and see if your filer
> is up to the task.  Are the reads coming from there as well, e.g. textures,
> models, etc.?  'hconfig -a' doesn't list any MANTRA env vars related to
> buffering, but that's something SESI might be able to implement if they
> haven't done so internally already, just not exposed the buffer size as an
> env var.
> -- Antoine
> On Sep 8, 2014, at 2:29 AM, Matt Estela <matt.estela at gmail.com> wrote:
> > Curious if anyone else has suffered this sort of thing;
> >
> > We sent a Mantra job to the farm, relatively fast per-frame times, but
> deep
> > exr's with lots of layers meaning it came in at around 120mb per frame.
> > When this was rendered by lots of farm machines at once (like 200 or
> more),
> > this was enough to bring the filer to its knees, lots of fingers pointed,
> > sad sysadmins.
> >
> > I know we've rendered stuff heavier than this before with PRman. Turns
> out
> > this problem had been identified and fixed a while ago, in that our Maya
> > PRman jobs automatically render to the local scratch disk locally per
> farm
> > machine, then copy the frame to the filer when done, keeping the network
> > happy.
> >
> > Anyone implemented such a thing in Mantra? Or know of mystery Mantra
> > options to force it to keep its network write operations to a minimum?
> >
> >
> > -matt
> > _______________________________________________
> > Sidefx-houdini-list mailing list
> > Sidefx-houdini-list at sidefx.com
> > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list

More information about the Sidefx-houdini-list mailing list