Hi everyone,

A simple poll: what render queue is your facility using? I'm trying to get a sense of what the Houdini community at large is using. Please include what platform, e.g. Linux, Windows, etc. Bonus question: how many procs in your render farm?

Thanks!

– Antoine

We're using Rush on Linux, around 40 procs.

– Antoine

Antoine Durr
We're using Rush on Linux, around 40 procs.

– Antoine

Might be worth adding if those are hardware procs or cores or what. So hard to tell these days. Like 40 = 5x8 core machines. or that could mean 40x8 core machines.

In the past: drQueue, Deadline, at the moment custom, based on python. 10x4.

cheers
kuba

Wolfwood

Antoine Durr
We're using Rush on Linux, around 40 procs.

– Antoine

Might be worth adding if those are hardware procs or cores or what. So hard to tell these days. Like 40 = 5x8 core machines. or that could mean 40x8 core machines.

Good point. I was using cores and procs interchangeably. It's 4x8 + a few 1x4's.

– Antoine

Rush here. Tested mantra on 100-200 cpus. Investigation of comprehensive RUSH/python/ROPs interface for managing Rush dependencies was favorable, (but not fully implemented or stress tested).

Race (Rush related) at DD is still the best system I've ever used. There was a grid engine based thing at a behemoth animation studio I worked at in L.A. that managed thousands of nodes ok, but the user interface was awful compared to race.

We looked at Temerity recently. That's more than just a renderfarm, but had some nice user interface tools for the renderfarm side of things.

Race (DD proprietary), and Queue (R+H proprietary), and a little work with Pixar's Alfred.

Alfred was used to manage multihost interactive rendering only, on a small pool of procs (~50) - not used for offline rendering to disk. We used it to reserve cpus for “mantra -H” and for PrMan's netrender. It's quite a setup and has a dense administration toolset but it does have some pretty good response times - something needed for multihost network rendering. It does cost some money though; but if you have MTOR, you probably have Alfred hanging around. The Alfred ROP in H9 currently needs to be customized to minimize wait-times for interactive rendering, though; it's currently setup to submit jobs intended for offline rendering.

jason_iversen
The Alfred ROP in H9 currently needs to be customized to minimize wait-times for interactive rendering, though; it's currently setup to submit jobs intended for offline rendering.

Jason: would you mind elaborating on this customization?
(I'm just getting started with the Alfred ROP…) thanks!

We are using irush on over 1000 cores running suse, not including the xserves for comp renders.

Use Alfred here to handle Houdini and all other jobs (Maya, Nuke, RealFlow, etc.). And, it was simply because we had Alfserver nodes around and it's really easy to script.
~140 procs.

Chris