Distributed Rendering takes ages to get data from nodes

[Dippndots]

This is probably an easy fix (hopefully), and I think its to do with network bandwidth, but I didn't have enough time to troubleshoot before I left the office. Basically I have 26 nodes rendering a 5k image; each did the required 5 or 6 passes in about 8min, but it took nearly an hour for all the data to get to the master machine and finish the render.

Should I reduce the max number of pixels transported or increase the time between sends? Or both?

[Nekrobul]

This is probably an easy fix (hopefully), and I think its to do with network bandwidth, but I didn't have enough time to troubleshoot before I left the office. Basically I have 26 nodes rendering a 5k image; each did the required 5 or 6 passes in about 8min, but it took nearly an hour for all the data to get to the master machine and finish the render.

Should I reduce the max number of pixels transported or increase the time between sends? Or both?

Try this numbers. 3 sec synchronisation interval Max pixels 50000. Thise worked out for me with limited VPN connection.

[Dippndots]

Thanks I'll give that a go on Monday. Perhaps I should've mentioned all the machine are on the same local network, if that makes any difference?

[Dippndots]

This worked pretty well, down to 27 minutes so only about ~17 minutes of it grabbing data from other nodes. Though I'm having another issue with all my render elements that might be prolonging this. I'll make a separate post for it.

[Nekrobul]

This worked pretty well, down to 27 minutes so only about ~17 minutes of it grabbing data from other nodes. Though I'm having another issue with all my render elements that might be prolonging this. I'll make a separate post for it.

27 minutes is quet long for single image. What kind of connetion do you have?

[Dippndots]

Well about 10 minutes of that 27 was rendering. All the machines are talking to each other at a minimum of 1Gbps

[Nekrobul]

Well about 10 minutes of that 27 was rendering. All the machines are talking to each other at a minimum of 1Gbps

This is realy realy wierd. Actyaly the data should be dumped while it renders and would not take any aditional time to dump it after render finishes.

[Dippndots]

Yeah, this was how 1.1 was working for us, I just thought it was one of the changes to distributed rendering in 1.2 or 1.3.

Just to be clear this is how the render happens: The master machine renders 5 passes (the total required passes are 125, so I'm assuming that it knows the other 25 machines will do 5 passes as well), another window pops up saying collecting data from slaves, at this point, the master machine stops rendering (cpu usage drops down under 20%). Then for the rest of the time, it is just collecting data from the slaves. All 26 machines are the same, so they all finish their 5 passes around the same time.

This after render dumping might be tied into my other problem with distributed rendering via backburner:
https://forum.corona-renderer.com/index.php/topic,10297.0.html