By: RichardC (tich.delete@this.pobox.com), January 25, 2017 6:30 am
Room: Moderated Discussions
RichardC (tich.delete@this.pobox.com) on January 25, 2017 4:26 am wrote:
> >And that their parallelization method is separate
> > frames. While the average network utilization is low, the peak requirements are incredibly high.
Now while the render farm would seem to have very low bandwidth requirement, you certainly
*do* need high bandwidth for serving full-resolution uncompressed video at 24 fps, e.g. for
review and editing. With about 30MB/frame at 24fps that's 720MB/sec, which is getting close to
the limit even for 10Gbit ether. Making that work for a lot of people working simultaneously,
without too many bottlenecks and glitches, is definitely a networking challenge. But it's
orthogonal to the design of the render-farm hardware. ISTM the render-farm is low-bandwidth
high-compute, and the storage infrastructure needs to be high-bandwidth low-compute.
> >And that their parallelization method is separate
> > frames. While the average network utilization is low, the peak requirements are incredibly high.
Now while the render farm would seem to have very low bandwidth requirement, you certainly
*do* need high bandwidth for serving full-resolution uncompressed video at 24 fps, e.g. for
review and editing. With about 30MB/frame at 24fps that's 720MB/sec, which is getting close to
the limit even for 10Gbit ether. Making that work for a lot of people working simultaneously,
without too many bottlenecks and glitches, is definitely a networking challenge. But it's
orthogonal to the design of the render-farm hardware. ISTM the render-farm is low-bandwidth
high-compute, and the storage infrastructure needs to be high-bandwidth low-compute.