how to customize number of core/memory to use per slave ?

Deadline
1

Hi, i have couple of question :

  1. i want to sumbit project and i want to assign it to multiple slave. So In the submit window in the machine list i can pick multiple slave to work with, and i can set number of thread and max of ram. But this threads/memory setup is generic setup for all the slave i want to use. In fact each of the slave has different thread and max memory i want to use. How to customize this so i can set up thread/mem per slave ?

  2. I’m using nuke. The subsmission window and the submit window give me no option to add command line argument. like in nuke i have -X (node) where node is the ‘write’ node i want to execute. How to deal with this ?

  3. is it valid if i create the output folder (for output storage) inside the deadlinerepository folder? or has be separated and be mounted separetely ? I think both can work, but which is the ideal solution? The documentation didn’t mention about this rule. what the pros and cons?

  4. in case where there are 2 slaves , each of them has 2 network interface (wireless and cable) so one computer has 2 unique IP addresses . Both the interface are active and working properly. My cable interface can tansfer much faster than the wireless. So how to tell the Deadline which network interface to establish connection ? is it possible ? coz i don’t want it to pick up the wireless which is slower.

Thanx.

2

Hi!

  1. Setting the thread count and max memory to 0 will allow Nuke to automatically figure out the appropriate numbers.

  2. In Deadline 5, there will be an option to use NukeX for rendering instead of Nuke. The release is only a few days away!

  3. We would recommend having a separate location for output. This keeps the repository clean and ensures that Deadline doesn’t accidentally touch your output images. For smaller farms, having both the repository and output on the same server is fine, but as a farm grows, it is a good idea to have them on separate servers so that the saving of output doesn’t affect the repository machine’s performance.

  4. There is no explicit option to control this in Deadline. However, Deadline would use the same network interface that Windows Explorer or any other application on your system would use.

Hope this helps!

  • Ryan
3

hi, that’s the problem with my question no.1 , if i leave it to 0, the rendering proccess is very slow. For example, i render inside nuke and it took about 1min16sec, but when i submit to deadline (test by using only 1 slave quad core) , it took more than 5 minutes ???
i check the cpu activity to see how all cores has been involved and i saw there’s very2 low cpu activity while rendering via deadline. Is there any way i can boost to force all cores to use their maximum power ? As comparison, when i use qmaster (apple ) , i can really boost the power. I can see in cpu monitoring tool in my mac, all the cpu cores got very high proccessing activity. so i using single slave (using the same workstation where i submit the job), i can get the speed of 150% so in my above case , i can get around 45 seconds or less.

Any idea ? did i miss something ? any parameter need to be set ?

Btw , it’s a fantastic news about Nuke X. I can’t wait :slight_smile:

Thanx

4

Is that per frame, or is that how long it takes to render the entire sequence? A Deadline job is normally split into multiple tasks, and each task can consist of one or more frames. Each task inherits a bit of overhead because Nuke needs to be launched for each task, so if you are only testing on a single machine, and you’re splitting up the job into multiple tasks, you will naturally see the job take longer than a local Nuke render. Of course, as you add more slaves, then you have tasks being worked on simultaneously, which will speed up the overall job time. If you know ahead of time that your frames will only take seconds to complete, then increasing the Frames Per Task to 5, 10, or 20 will help reduce overhead as well.

To truly test a Deadline’s job speed on a single machine versus a local render time, you have to set the Frames Per Task to equal the number of frames in your sequence (ie: if rendering frames 1-100, set the Frames Per Task to be 100). This will create a job with a single task, and will render all the frames in one shot.

Hope that clears things up!

Cheers,

  • Ryan