Distributed Rendering

From BRL-CAD

Distributed Rendering

In order to create a distributed render across many machines using BRL-CAD, one first needs a 'host' that will serve as the rally point to where all rendering machines will go to get information needed for the render.

Typically the host machine would have all the information needed in order to carry out the distributed render however, that currently doesn't work right now.

For the moment, all machines must have a copy of the information needed for a render to be completed. So for example:

'Server' is the host machine in which all computers will report to in order to get their jobs, and return their jobs to. 'comp1' and 'comp2' are machines that will connect to 'Server' in order to carry out the rendering job.

So in order for the whole process to get started, one would have to access 'Server' and run the remrt command.

For Example:

$ remrt -s100 Picture.g objects.c

or: remrt (- options) (.g file) (.c objects in .g file)

would start a server that is going to render a 100 x 100 picture using geometry from Picture.g with objects.c loaded.

Then for comp1 and comp2 you would run:

$ rtsrv Server 4446

or: rtsrv (server's name) (listening port, which defaults to 4446)

to connect to Server, and begin the rendering job. 'Server' will assign work to 'comp1' and 'comp2' and they will start rendering their lines for the picture. Typically the workload is that each connected computer will render a single line of the picture, so comp1 will render the first line, and comp 2 will render the second, and will continue rendering until all lines are processed.

The final rendered picture is located on 'Server' in the .pix format.


Advanced Commands

HOWEVER, this all is pie-in-the-sky simplicity and in reality much more work has to been done in order for proper distributed rendering to occur.

Currently, remrt needs many extra values given to it in order for it to work correctly. This line is an example of the actual information needed in order for a proper server to be set up:

$ remrt -M -s4096 -p65 -o Output.pix shape.g object.c 2>> Log.log <<EOF
> viewsize 1.298471982398
> orientation 1.1238190283 2.39028130938 3.0129381098309 4.120938019823
> eye_pt 2.192310238018 3.120398109283098 4.108312083108
> start 0; clean;
> end;
> EOF

And the rtsrv has to be run like this:

$ rtsrv -d Server 4446

And that will connect your current computer to the server to start rendering the picture.

One important thing to remember is the .remtrc file, placed on the server, which holds information on which rendering computers do what and when.

.remtrc

host computer1.brlcad.org always cd /tmp
host computer2.brlcad.org always cd /geometry
host busyserver.brlcad.org night cd /tmp

This shows 3 computers that will be used in the render, computer1 and computer2 will render 'always' in their respective directories where the geometry is, and busyserver will only render at 'night' (6pm - 8am), keeping busyserver from becoming even more busy during normal working hours.