[REQ] Improved network code

[kadajawi]

At least I have a big problem with network rendering. I always get

Code: Select all

Receiving frame from 192.168.0.2:
        num_samples: 1120000.000000
        width: 3204
        height: 2336
ClientProxy: MySocketExcep: read failed, error: [unknown]
Dropping connection to slave.

That means... sometimes I get it, sometimes I don't. Sometimes it runs for hours without a problem, but then (usually when I am not at home) it fails after 2 minutes (as soon as I'm out of the house). When the server computer is busy I also have this problem. I don't know what it is, but it's extremely annoying since I have to restart the rendering on that computer.

Maybe you could make it so that it retrys to send the data when it fails? I guess that shouldn't be too hard to do, maybe also with a setting in the inifile to choose how many trys you give. Another thing may be an option so that you can switch off the server and it still continues to render, and when you start the server again the slave sends what it has done meanwhile.

Maybe it would also be more robust to render when the slave actually saves the samples as igi file, which is then merged by the server. But I think just restarting the network connection between master and slave should be enough.

[OnoSendai]

I think your router sucks

[kadajawi]

I'm not using a router, the computers are connected via a crossover ethernet cable. And everything else works perfectly fine.

I think a retry feature would also be helpful for people who render over the internet, in case something fails there.

Btw. when I reduced the multiplicator of the server the connection also failed... or when I render another scene with a higher priority than the master client, or when I burn a DVD and the harddisc is constantly working... all that makes it fail.

[OnoSendai]

The render slave should try and reconnect automatically, tho right?

[kadajawi]

Hm, it should? I would like that, but all I get on the slave is:

Code: Select all

Uploading frame:
Frame size: 85.654 MB
Num Samples: 1120000.00000
Render thread terminating.
Render thread terminating.
MySocketExcep: write failed.  Error code == WSAEWOULDBLOCK

After that it just sits there and does nothing... can wait for hours.

[OnoSendai]

yes.. but then what?

[kadajawi]

Nothing. I just waits and does nothing, CPU is idling, ...

[psor]

@kadajawi

Quick question, which Indigo version do you use right now? ;o))



take care
psor

[kadajawi]

1.0.5.2. Couldn't migrate to 1.0.6 yet since the current scenes were started with 1.0.5.2. Sorry for not mentioning that... 1.0.6 is pretty new.

[OnoSendai]

What does the client print out after it gets disconnected?

[kadajawi]

Nothing. What I quoted is all I get. After that nothing at all happens. The client is running on OSX (I guess I keep forgetting the important bits? ) with Crossover (Wine). The client is also still running, I have to press strg+c.

I will test now what happens when I run the server on the Mac.

[OnoSendai]

That's odd, because the client is supposed to loop and try and connect again.

[kadajawi]

Ok, I have tested running the server on the mac now, with even weirder results. The PC client sends the packages flawlessly to the Mac server, but the Mac client fails to send its data to itself. I will try 127.0.0.1 now...

I think the problems are related to Wine/Crossover, but I recall other users reporting that it works for them.

Tell me if there is anything I shall test for you.

Edit: Ok, 127.0.0.1 doesn't work either.

[kadajawi]

Ok, what works for me now is running a working master on the Mac and the client on the PC. Unfortunately the Mac isn't connected to the internet, so when I try to let computers over the internet join that would be a problem. But for now this works fine

I still think it is a problem with the master not being able to process the incoming data fast enough (e.g. CPU too slow, too little priority, at the moment busy doing other things... (like waiting for the HDD... stupid Windows)).