When rendering to a shared network resource, will V-Ray ever write blocks continously during render?

[Fredrik Averpil]

I was asked to post this question here by our IT consultants...

During mayabatch with V-Ray or vray standalone rendering, and when rendering directly to a network share, I am under the impression that V-Ray would store data in memory and on local disk during rendering and then not until when the render is 100% completed, V-Ray dumps the results (images, irradiance maps etc) onto the network share in "one go". Is this correct?

[yolov]

V-Ray usually stores the image in memory and at the end of the frame writes it to disk - be it on a local or network location, however it's set up.
Additionally there's the option to write directly to disk without storing it in memory - it's only for the bucket render, though. It only works for .vrimg and multi-channel EXR files. As each bucket completes - it's directly stored on disk. The 'memory buffer' options in the common tab are used to switch between writing to disk after each bucket, or storing the image in memory and writing to disk at frame end.

The setup is described here: https://docs.chaosgroup.com/display/...y+Large+Images

The resumable rendering features does something very similar, here's a link: https://docs.chaosgroup.com/display/...able+Rendering

[Fredrik Averpil]

Great, thanks.

[Fredrik Averpil]

But what about if RAM is not enough, doesn't V-Ray start swapping to disk then?
I assume such temp data which is written during render never is written to the render directory, right?

[yolov]

Although I can't be 100% sure, I can take an educated guess that swapping will be done locally on each slave/master as RAM is maxed out.
Then, when talking about memory - it's good to specify if we're talking about the memory required by the image buffer (that keeps the pixels for the final image) or anything else related to memory in V-Ray?
If you're doing DR, the slaves hold some information about the image that they regularly send back to the master - the master node puts it back together. The slaves will (if it comes to that) swap locally as they're a local process to the machine where they are running.

The image is kept in a buffer and if VRay can't allocate it at the start of the rendering - it will fail.
If you're using the option to write the image directly to disc (with the bucket sampler and with disabling the image memory buffer) - then memory is not an issue.

Then, the image buffer aside, other potential memory issues can come from the scene itself (geometry, textures and all sorts of assets) - which is handled pretty efficiently with the bucket sampler (even when using the image memory buffer).
Ideally, you wouldn't want to have a network setup where some of your DR nodes have 128G of ram and others have 32G when doing large scenes for obvious reasons.

[Fredrik Averpil]

The background to my questions is that we are seeing increased writing (to file server) activity during rendering in our network. I would like to exclude the possibility that V-Ray adds to this network traffic, as I'm assuming that V-Ray does not write anything to the render folder during rendering. This is regular commandline mayabatch rendering and not DR.

EDIT: I mean, except for dumping the final frames onto the network share when rendering is 100% done, of course.
I know that for example mental ray used to open up a stream and continuously write to the image files in the render folder during rendering. I believe Nuke may also be doing this..
If it turns out that some software is writing continuously over the network, we might want to render to local disk instead, and then finally copy the results onto the file server as a post process when the rendering is 100% done. And right now we are investigating how the DCC apps we are using are behaving in regards to this.

[smbell]

yeah, nuke is definitely writing as it renders!

[yolov]

You can try with VRay Standalone outside maya to replicate it with the same machines - that way you can be sure if VRay has something to do with it.
If you could tell me more details - I can test this here on my end. What's the mayabatch setup?