Server- Workstation and the rest

[Amen]

Hi,

 

I am having a discussion with our network admin. He is worried about the traffic in the network and of the storage usage with V-Ray.

So we are trying to find a good solution for 4 workplaces and a couple render drones.

 

My first idea is, to get four machines with a dual processor 10-core and 64-128gb ram, 780 gtx 6gb + two of those without a good graphic card (+ maybe some existing machines) purely for rendering . Data is on a server for the whole office (around 20 people). This setup cost around 25k-30k €

That should work for 2-3 years if the workload stays the same.

 

His solution is, to have a rack with similarly build servers + dedicated storage. We would then use Pcoip to connect to those servers. He likes this because most of the traffic would stay with the servers in the same rack. He thinks, he can get that for around 5-10k less.

 

The classical solution... Workstation + drones in a server rack, he did not like that to much, because he thinks the traffic gets to much when distributed rendering.

 

My questions:

 

1. Has anybody used Pcoip with Max and V-Ray?

 

2. Does V-Ray work on a virtual machine? And do I lose performance? Is the setup easier with a virtual machine (using plugins... licences)?

 

3. How much traffic does V-Ray produce?

 

4. What would be an ideal storage solution for V-Ray?

 

5. What would be the perfect setup for Max + V-Ray?

 

Thx for your help.

 

Bye,

 

Jan

 

I posted this on Chaosgroup forums:

http://forums.chaosgroup.com/showthread.php?78197-Server-Workstation-and-the-rest

Edited May 13, 2014 by Amen

[numerobis]

I only used Remote Desktop or Teamviewer to work over network and this always produces some lag even over 1 Gbit network if i remember correctly. I'm not sure how PCoIP is comparing to that. Nvidia is pushing such solutions, but i'm not sure how this would work with 1 GBit.

 

But my question would be how you plan to connect the nodes to the server... infinyband? 10Gbit network? A separated 1 GBit network/switch? Otherwise the traffic would still be in the network without any improvement.

 

And i think the traffic produced by vray is fairly low and only raises when the scene gets loaded.

 

If this is really a problem i would invest the money into a switch with 10GBit connection or maybe several 1Gbit cards combined (teaming) to the server because this should be the bottleneck if everyone is connecting to the server.

 

I would go with normal workstations.

 

Do you need to buy within the next time or is it possible to wait a few months? Haswell-E is coming - but it looks like the dual Xeons (Haswell-EP) are released later in september or october. But i think i would wait for the new platform (new socket, DDR4, upgradable with Broadwell-EP) if you don't need it tomorrow.

Edited May 12, 2014 by numerobis

[Amen]

I do not think vray produces to much traffic, but I need some real data to prove this to our network admin. Does anybody have that?

 

The discussion I have with our admin is, to have a few dedicated 3d machines and use those in the office that are powerful enough as render nodes.

 

The network has a few switches and is based on 1GBit. Therefore 3d stuff and other office stuff use the same network.

Our admin would like to have the network separated in 3d and the rest. His ideal is having everything in one server rack. That is where PCoIP comes into play. I like that solution, too. It seems quite elegant, but only if the lag is not noticeable. Does anybody have experience with pcoip???

 

The Haswell E or EP look really nice, I think I can wait that long.

 

Until than I keep looking for an ideal solution. Very efficient with maintenance (maybe cloning and virtual machines work well?)