Scheduling Algorithms

Notes

• This is 5.3
• The average waiting time is as you would expect
  • Add up all of the time processes spend waiting
  • Divide by the number of processes
• This is a simple measure of performance and consistency
• Notation Pn-t
  • n is the process number
  • t is the time for a burst
• So if we are scheduling P1-24, P2-3, P3-3 we would take 30 seconds for all three to run.
• We are currently assuming no preemption
• First Come First Served - no preemption (FCFS) : the first process in the queue runs for its burst then the next, then the next, ...
  • Queue: P1-24, P2-3, P3-3
    • Wait times: P1:0, P2: 24, P3:27
    • Average 51/3 = 17
  • Queue: P2-3, P3-3, P1-24
    • Wait times: P1:6, P2: 0, P3:3
    • Average 9/3 = 3
  • The system tends to degenerate into the first example
    • This is called the convoy effect
  • FCFS is good because
    • Easy to implement
    • Fair in that every process gets an equal chance to run
    • No starvation
    • Very low cost algorithm
    • Very predictable
  • But it is not
    • Good for interactive systems.
    • And has a high average waiting time.
• Shortest-job first - no preemption
  • Select the shortest job and let it run first.
  • To do this, we need to estimate the next burst time
  • This can be done with the algebraic average
    • But it is better with an exponential average
    • $\tau_{n+1} = \alpha t_n + (1- \alpha) \alpha t_{n-1} + ... + (1-\alpha)^i \alpha t_{n-i} + ... + (1-\alpha)^{n+1}t_0$
    • $0 \le \alpha \le 1$
    • And $\tau_{n+1} = \alpha t_n + (1-\alpha)\tau_n$
    • This considers all bursts, but places more weight on the last burst.
    • See figure 5.4
  • Advantages
    • Minimum waiting time, but for short processes
    • High Throughput
    • Short turnaround
  • But
    • Starvation
    • Unfair
    • Higher overhead.
• Round Robin
  • FCFS with preemption
  • define a time quantum or time slice.
    • This is how long a process can run without interruption
  • The size of the quantum is important
    • Small supports lower average wait time
    • But also means that more time is spent in os operations (less progress)
    • Larger means more progress
    • But waiting time goes up, and it does not feel as interactive.
  • The quantum should be large compared to the time for a context switch
  • A good rule of thumb is that 80% of the bursts should be less than the quantum.