Slow pickup on clients

[markeljfp]

Hi

v8.3.0

I am finding that jobs are picking up really slowly on eligible render nodes.

e.g.: I have six nodes with four threads each waiting to pick up comp jobs, I'm lucky if two threads are active at any time., the other 22 remain idle.

What is the best way to get them to occupy the threads I have set up? I have

• checked that the jobs are assigned,
• that the nodes have the correct Nuke version,
• used the option to rescan for installed software.
• reinstalled the rrClient on a couple

What else could be holding them up?

Hope you can help!

[RR support, Schoenberger]

Hi


If you open rrControl, tab Job Log and select the Job Log, what is the
time per segment? (last column)
If it is very low, it could be this issue:
http://www.royalrender.de/help8/index.html?SequenceDivide.html

In this case you should increase at least the Sequence Divide MAX (Max
is more important than Min. The more jobs you send, the larger the
frame segments send to clients).


Notes:
You can see the average time for all jobs including 3D:
Please open your rrServer app log file RR/sub/log/server...txt
There is a log entry every 30 minutes:
"Segments done (since last half hour):105 Average time per
segment:34.00 m.s"



regards,
Holger Schönberger

Please use the rrKnights Tavern or
our support system for new questions

[RR, Schoenberger]

PS:
You might want to lower the client connection delay in rrConfig/Client Global a bit if it is that issue with the short segment times.
This might help a bit.

regards,
Holger Schönberger

Please use the rrKnights Tavern

or our support system for new questions.

[markeljfp]

Hi Holger

The times are pretty quick, admittedly, at 62 segments and 01.24 m.s average.

What I don't understand is why, when our farm is otherwise full, we have 5 jobs of 306fr waiting and 6 nodes not picking the jobs up. Barely on threadA, and definitely not on the other three threads. I've now also used the debug to check the decision making and it looked good. e.g.:

Jf-Ws-D017-Mull:
  {NA5 }     able to send   |Priority-Scores: Exclusive000.Privileged001.AvrA391.AvrB391.C6368
         (Info: Client rendered job for 00.00m.s, got job 0 times, crashed 0 times)

It _feels_ like there is a delay between sending packets to a node, but it's just too high. We can't have it at zero, because that causes other issues when copying files to a local drive (especially when there are multiple instances of the same job on the farm).

[RR support, Schoenberger]

Hi Mark

> The times are pretty quick, admittedly, at 62 segments and 01.24 m.s
> average.

If there would be a delay of 10 second, then 8 job threads can render
this comp.

There is a delay between different threads of a client. If a job
threads of a client gets a job, the other ones are lin wait state. You
can see that in rrClientWatch and rrControl.

Afaik the option to have a delay between sending 2 jobs at the
rrServer is not active any more, but I can take a look if there is a
config setting to enable it.

And at last there is the client to server delay that I already
mentioned in my last email.

[RR, Schoenberger]

And another setting to check which only affects clients that are idle.
rrConfig, tab Clients Global, right side.
The defer mode should at least be set to 15, better 20-25 in your case.


regards,

Holger Schönberger

Please use the rrKnights Tavern

or our support system for new questions.

[RR, Schoenberger]

Hi

Another note:

There is a setting in rrConfig, tab Jobs, right side
"Wait [0] seconds between sending jobs (NOT recommended)"
Have you changed that value?