Round-robin with CPU-bound Processes


Suppose that processes p1, p2, p3, p4 are all cpu-bound (no I/O).

Round-robin on these cpu-bound processes is not really a good idea (why?).

|--------|--------|--------|--------|--------|--------|--------|--------|--------|--------|--------|--------|  (short time slices)
    p1   |   p2   |   p3   |   p4   |   p1   |   p2   |   p3       p4       p1       p2       p3       p4 
         |        |        |        |        |        |
         |        |        |        |        |        |
      context  context  context  context  context  context
      switch   switch   switch   switch   switch   switch


Remember that RR is good for response time, but has bad turnaround time.
For these cpu-bound processes, turnaround time is probably more important than response time.
It is probably not important to p2 that, with RR, it starts up sooner (but ends much later).

For these cpu-bound processes, RR is causing too many unneeded context switches and RR is
causing overly long turnaround times.

Either run each process to completion (essentially, FCFS) or do RR with a very long time-slices.
That has two advantages over RR with small time-slices. Most of the processes get a better
turnaround time, and we do not waste so much time on multiple, unneeded context switches.


|-------------------------|-------------------------|-------------------------|-------------------------| (long time slices)
    p1                    |   p2                    |  p3                     |   p4
                          |                         |                         |
                       context                   context                   context
                       switch                    switch                    switch