NSOperation for multiple small pipelines?

FROM : David Carlisle
DATE : Wed Nov 07 22:11:07 2007

If I have a program that can create multiple small pipelines of 
arbitrary length, would NSOperation be effective for distributing the 
processes among several threads?  A useful answer for me would be 
"Yes, just study the documentation" or "no, it won't solve the problem."

If I have one pipeline, A-B-C, then process A would put data into a 
queue for process B, which would put data into a queue for process C. 
If only one thread were available, I would want it to completely 
execute process A, then completely execute process B, etc.  If three 
threads were available, I would want each process to be assigned to a 
different thread.  If one of the processes crashes, then it needs to 
cause the other processes in that pipeline to gracefully terminate.

Then lets say I have three pipelines, A-B-C, I-J-K and X-Y-Z.  If 
three threads were available, then they could possibly be assigned to 
processes A, B and C, or possibly to A, I, and X.  If more threads 
were available then they should be assigned to processes that are 
likely to have available data.  In the case where processes A, I and X 
are active, then processes likely to have data would be processes B, J 
and Y.

I'm assuming from what I read before Leopard was released that 
NSOperation could be informed that process B was dependent on process 
A.  So does that mean that NSOperation will not begin process B until 
after process A has finished?  Or will NSOperation be able to begin 
process B after process A has begun and allow them both to operate 
concurrently?  Are there implications for how I have to put locks on 
the queues?

My present implementation just causes a thread to be created for every 
process, but now I understand that the new way to approach the problem 
is to use NSOperation and allow it to decide how many threads should 
actually be created.

Any thoughts?