RR Scheduling Algorithm

Advantages:
-- More interactive for users
    (quicker user response time?)
-- No starvation
-- Less overhead (less complex)
-- Easy to implement

Disadvantages:
-- Longer overall wait time
-- Longer overall turnaround time
-- Does not handle priorities

Impact of selecting the time slice?
-- Too big: FCFS
-- Too small: Way too many context switches
              (too much overhead)


Preemptive Priority Scheduling

Given: two processes

       P1 is CPU-bound

       P2 is I/O-bound

Which of these processes should be
  assigned the higher priority?

 Goal: Minimize turnaround time


P1 should have higher priority?  Why?
 -- Faster turnaround time, because
    it doesn't have to wait for I/O?

P2 should have higher priority?  Why?
 -- Get CPU first, quickly get out
     of CPU to doing I/O, allowing
     other processes (P1) to do their work

Example:
P1 (CPU-bound) requires 300ms of CPU
P2 (I/O-bound) requires 80ms of CPU
               and three 60ms I/O operations
               that fire at 20ms intervals
RR with time slice of 100ms

What are the turnaround times?

If P1 has higher priority,
  P1 takes 340ms and P2 takes 440ms
    (average is 390ms)

If P2 has higher priority,
  P1 takes 380ms, but P2 takes only 260ms.
    (average is 320ms)

===============================================

Exponential averaging (see lecture notes):

What if alpha = 0?  Prediction is constant;
              actual history has no effect

What if alpha = 1?  Prediction is purely
               based on actual history
       (i.e. last measured CPU burst time)

Tau0 = 10   alpha = 0.5   t0 = 6ms

Recalculate using alpha = 0.25 and 0.75