DR - send to each node or shared folder - what's faster?

[Micha_cg]

Hi,

I ask me, allow one of this modes a speed up of the rendering? The tool tip tell, that the shared folder mode is faster. But if I think about it, than I found, that the scene and the textures must transfered to each slave in both cases, so where could be the speedup? Start's the rendering at the master earlier in shared folder mode and but the slaves need the same time?

I like at the "shared folder", that I can easy pick all textures from there to my project directory.

Best-
Micha

[dalomar_cg]

Re: DR - send to each node or shared folder - what's faster?

First of all, a shared folder is one that should be in the same absolute location for all nodes. If its just the same folder location, and you have to copy them all, then you're really faking the shared folder.

AFA why transferring is slower is because of the transfer being based from the host computer as opposed to a file server. Sending all of those maps through the network will take a little while from a standard computer because it isn't set up to send out mass requests for information. Typically shared folders are on a file server with a RAID array hooked up, so the transfer from a dedicated server to a node will probably be faster. Your right in the sense that the nodes have to get the files from somewhere, but what type of computer will typically matter. In the case of your home DR setup I'm not sure you'd see the speed up, since your not working with a dedicated file server, but a full fledged company probably would.

Also, I'm not sure, but the transferring the files may also require that a temporary file will be written as opposed to reading the file directly from the location. If thats the case, then a shared folder would definitely be faster. I'm not sure on this though, so don't quote me on it.

[Micha_cg]

Re: DR - send to each node or shared folder - what's faster?

Thanks Damien. My network shared folder is a folder at the master engine, the network is a 1GBit network and the HDDs are SATA HDDs. If I'm right informed, than the SATA standard allow 1.5GB/s or 3GB/s. So, the network could be the bottle neck. Or are the HHDs slower than the SATA standard? I'm not sure.

So, if I understand anything right, for my hardware it dosn't matter, what mode I use.

[dalomar_cg]

Re: DR - send to each node or shared folder - what's faster?

I'm by no means a network guy and for hardware I haven't been keeping up with things. I'd say do what works for you. The only way these things really begin to separate themselves is when your talking about 5-10 nodes. Since everything is all on one machine you'll probably won't seen a difference.

[Luke72]

Re: DR - send to each node or shared folder - what's faster?

The better Solution is a dedicated server, or the alternative is a litte nas, like lacie "network disk" where you can configure a static or dynamic tcp ip.
I work with a NAS, and its ok, but sometimes have problem of speed balancing of the render nodes on the local network. The spawner sometimes not work.
I work with simple windows network, configured in workgroup mode.
But maybe is better create a local domain?
for example, in a workgroup network is not possible ping a specific render node, to check if is busy, i whish this for the next release of the plug..
Sorry for my english :

[Micha_cg]

Re: DR - send to each node or shared folder - what's faster?

So, the question behind the speed difference is, what is faster - write the scene to the master (network) disk or to the slave disk? In my case, there should not be so much difference.

Sidenote: I found an other difference between both methods - a big advantage of the sharded folder is, that IM/LC cache files can be easy saved there and all slaves found it. I tested it, local saved cache files are not tranfered to the slaves. Or should this be done like with the textures maps, cache files send to each slave? Why not?