LAE

Note: This was originally posted by an inactive account. Content was preserved by moving under an admin account.

Originally posted by: sgerbig700

What is the maximum number of nodes a graph can handle before operating efficiency degrades? For example , a "Not Responding" from the server, taking minutes not seconds to START running a node or editing a node.

Note: This was originally posted by an inactive account. Content was preserved by moving under an admin account.

Originally posted by: stonysmith

>>>"before operating efficiency degrades"

That's a very tricky question. I've seen 16,000 node graphs that work just fine.

However, there is a thing that often causes some inefficiencies.
If you have two composites that are CLOCKED together, you should go inside the composites and connect the "InClock" and "OutClock" pins.

Imagine two composites of 10 nodes each. They are clocked together, but the InClock/OutClock options were not setup.

In that case, every node in Composite A is clocked to every node in Composite B
That's 100 clocks! As that number grows, the controller has to check more and more dependencies before it can do it's job.

If you connect the InClock and OutClock buttons inside of the composite, then you reduce it to only one clock that the controller has to check.. between the "last" node of composite A and the "first" node in composite B

Note: This was originally posted by an inactive account. Content was preserved by moving under an admin account.

Originally posted by: sgerbig700

... and what if clocks are not used at all? Should a graph with 20,000 nodes run fine.

Note: This was originally posted by an inactive account. Content was preserved by moving under an admin account.

Originally posted by: stonysmith

That's quite difficult to say. There's no restriction on the number of nodes... it becomes a question of how much "work" are you trying to perform?

If you're capturing the LOG.BRD files from the executions, you could go in an analyze the "interstitial" timing - how much time is spent running nodes vs how much time the controller spends deciding what to do next.

Note: This was originally posted by an inactive account. Content was preserved by moving under an admin account.

Originally posted by: stonysmith

Also.. what is the Thread Count set to on your FARM parameter?
You might be allowing too many nodes to start at once, all in parallel.

ls.brain.controller.farm=( "${ls.brain.server.host=localhost}:${ls.brain.serv er.port}:4" )

That number "4" there is the number of threads you allow for EACH graph that is running.
There is quite a bit of testing to do to find the "best" number, but think of this: the TOTAL nodes running should be no more than four times the number of CPUs that you have.

Example: 8 cpus * 4 = 32 threads (max)
If you have six graphs running with every one of them set to 16 threads (96 threads total) ... your server is over-loaded.

Note: the 4 threads per cpu is simply "rule of thumb" with a good bit of experience behind it. Your situation may produce a different need.
"Fastest" performance may be if you were to limit to only one task per cpu, but a lot of situations have shown me that the top end is four per cpu.