CPU Scheduling Tanenbaum Ch 2.4 Silberchatz and Galvin Ch 5 Basic Concepts • CPU - I/O Burst Cycle • CPU Burst Distribution 160 140 120 100 80 60 40 20 0 cs431-cotter 0 8 16 24 32 2 CPU Scheduling • Decision points: – – – – Process switch from running to waiting state Process switch from running to ready state Process switch from waiting to ready state Process terminates • Scheduling types – non-preemptive – preemptive cs431-cotter 3 Task Dispatcher • Functions: – Switching context – Switching to user mode – Jumping to PC location • The time needed to stop one process and start up another is known as dispatch latency. cs431-cotter 4 Categories of Scheduling Algorithms Batch Interactive Real time cs431-cotter Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 5 Scheduling Algorithm Goals Figure 2-39. Some goals of the scheduling algorithm under different circumstances. cs431-cotter 6 Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 Scheduling in Batch Systems First-come first-served Shortest job first Shortest remaining Time next cs431-cotter Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 7 Scheduling Algorithms (First-come, First-served) • Determine avg waiting time if: – (P1, P2, P3) or (P3, P2, P1) Process Burst Time P1 P2 P3 24 3 3 0 cs431-cotter 10 20 30 8 Scheduling Algorithms (First-come, First-served) • Determine avg waiting time if: – (P1, P2, P3) Process Burst Time P1 P2 P3 24 3 3 0 10 Waiting time: P1: 0 P2: 24 P3: 27 avg: 17 20 30 P1 ……………………………………………..P2…..P3……. cs431-cotter 9 Scheduling Algorithms (First-come, First-served) • Determine avg waiting time if: – (P3, P2, P1) Process Burst Time P1 P2 P3 24 3 3 0 10 Waiting time: P1: 6 P2: 3 P3: 0 avg: 3 20 30 P3….P2….P1………………………………………………... cs431-cotter 10 Scheduling Algorithms (Shortest Job First) • Schedule jobs based on their length. • May be preemptive – If a job is queued that is shorter than the remaining time on the current job, there is a switch. • May be non-preemptive. – Once a job has started, it works to its normal switch. cs431-cotter 11 Scheduling Algorithms (Shortest Job First) • Non-Preemptive / preemptive SJF Process P1 P2 P3 P4 0 cs431-cotter Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 10 20 30 12 Scheduling Algorithms (Shortest Job First) • Preemptive SJF Process P1 P2 P3 P4 Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 P1 | 0 10 20 30 P1.. cs431-cotter 13 Scheduling Algorithms (Shortest Job First) • Preemptive SJF Process P1 P2 P3 P4 P1 | Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 P2 | 0 10 20 30 P1..P2.. cs431-cotter 14 Scheduling Algorithms (Shortest Job First) • Preemptive SJF Process P1 P2 P3 P4 P1 | P2 | Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 P3 | 0 10 20 30 P1..P2..P3 cs431-cotter 15 Scheduling Algorithms (Shortest Job First) • Preemptive SJF Process P1 P2 P3 P4 P1 | P2 | Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 P3P4 | | 0 10 20 30 P1..P2..P3P2.. cs431-cotter 16 Scheduling Algorithms (Shortest Job First) • Preemptive SJF Process P1 P2 P3 P4 P1 | P2 | 0 Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 P3P4 | | 10 20 30 P1..P2..P3P2..P4……. cs431-cotter 17 Scheduling Algorithms (Shortest Job First) • Preemptive SJF Process P1 P2 P3 P4 P1 | P2 | 0 Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 P3P4 | | 10 20 30 P1..P2..P3P2..P4…….P1………. cs431-cotter 18 Scheduling Algorithms (Shortest Job First) • Non-preemptive SJF Process P1 P2 P3 P4 P1 | P2 | 0 Arr. Time Burst Time 0.0 7 2.0 4 4.0 1 5.0 4 P3P4 | | 10 20 30 P1…………..P3P2…….P4……. cs431-cotter 19 Scheduling Algorithms (Shortest Job First) • Determining the length of the jobs. • Use weighted average burst lengths n 1 * tn (1 ) * n 0 1 cs431-cotter 20 Scheduling Algorithms (Shortest Job First) 1 2 cs431-cotter 21 Scheduling in Interactive Systems cs431-cotter Round-robin scheduling Priority scheduling Multiple queues Shortest process next Guaranteed scheduling Lottery scheduling Fair-share scheduling Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 22 Round Robin Scheduling • Developed in response to time-sharing systems. • Each job is given control of the CPU for a short period - a time quantum. (In the range of 10-100 milliseconds.) • Control then passes to the next job (FIFO?) • Average waiting time dependent on job size, quantum size. cs431-cotter 23 Round-Robin Scheduling Figure 2-41. Round-robin scheduling. (a) The list of runnable processes. (b) The list of runnable processes after B uses up its quantum. cs431-cotter 24 Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 Round Robin Scheduling Process Burst Time P1 P2 P3 6 3 1 P4 7 0 cs431-cotter 10 All processes arrive at T=0 Vary Quantum time from 1 to 7. Calculate average turnaround time (to job completion). 20 30 25 Round Robin Scheduling Process Burst Time P1 P2 P3 6 3 1 P4 7 0 All processes arrive at T=0 Quantum Average Size Turnaround 1 11 10 20 30 P1P2P3P4P1P2P4P1P2P4P1P4P1P4P1P4P4 cs431-cotter 26 Round Robin Scheduling Process Burst Time P1 P2 P3 6 3 1 P4 7 0 All processes arrive at T=0 Quantum Average Size Turnaround 1 11 2 11.5 10 20 30 P1..P2..P3P4..P1..P2P4..P1..P4..P4 cs431-cotter 27 Round Robin Scheduling Process Burst Time P1 P2 P3 6 3 1 P4 7 0 All processes arrive at T=0 Quantum Average Size Turnaround 1 11 2 11.5 3 10.75 10 20 30 P1. . P2. . P3P4. . P1. . P4. . .P4 cs431-cotter 28 Round Robin Scheduling avg. turnaround 13 12 11 10 9 8 1 cs431-cotter 2 3 4 5 6 7 Quantum 29 Priority Scheduling • Each process (or class of processes) is given a priority. • Jobs are executed in priority order • Issue: If there are sufficient high priority jobs, lower priority jobs may never get scheduled. • One solution: Aging of processes. After a job has been in queue for a given period of time, raise its priority. cs431-cotter 30 Priority Scheduling Figure 2-42. A scheduling algorithm with four priority classes. cs431-cotter Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 31 Multi-level queue Scheduling • Partition the job space into distinct classes or queues – foreground / background – system / interactive / editing / batch / student – etc. • Independently assign queueing service disciplines • Assign queue priorities – Highest to lowest (empty one queue before starting next) – Divide time between queues • 80 / 20 • 30 / 20 / 20 / 20 / 10 cs431-cotter 32 Multilevel Feedback Queue Scheduling • Similar to Multilevel queue scheduling, but jobs are allowed to move between queues. • Avoids process starvation by allowing neglected jobs to move up to a higher queue. • Differentiate queues by (for example) quantum size – queue 1 = 8 ms – queue 2 = 16 ms – queue 3 = FCFS cs431-cotter 33 Thread Scheduling (1) Figure 2-43. (a) Possible scheduling of user-level threads with a 50-msec process quantum and threads that run 5 msec per cs431-cotter 34 CPU burst. Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 Thread Scheduling (2) Figure 2-43. (b) Possible scheduling of kernel-level threads with the same characteristics as (a). cs431-cotter 35 Tanenbaum, Modern Operating Systems 3 e, (c) 2008 Prentice-Hall, Inc. All rights reserved. 0-13-6006639 Multiple Processor Scheduling • CPU scheduling more complex when multiple CPUs are available • Homogeneous processors within a multiprocessor are usual (heterogeneous processors found in distributed systems). • Load sharing used with SMP. – Single queue, multiple servers. • Asymmetric Multiprocessing (AMP) – Assign roles (system control, application processing) – Avoids system data sharing cs431-cotter 36 Benefits of Multiprocessors Process Scheduling – Process inter-arrival time variation can be characterized by coefficient of variation = s / s • standard deviation of interarrival time / mean service time • ratio of 1 = exponential distribution. (your mileage may vary…) – Differences between scheduling algorithms become much less important in multiprocessor systems RR to FCFS thruput ratio Single processor cs431-cotter Dual processor Coefficient of Variation 37 Benefits of Multiprocessors Thread Scheduling • Allows true parallel processing within an application • However, if there is significant interaction among threads, small differences in thread management & scheduling can have big results. • General Approaches used – Load Sharing - (not necessarily load balancing) – Gang Scheduling - Schedule related threads together – Dynamic Scheduling - allow # of threads to vary cs431-cotter 38 Real-Time Scheduling • System Response Classifications – Hard real-time System - requires response guarantees – Soft real-time System – response not guaranteed • Dispatch Latency Interrupt processing Response interval Dispatch latency conflicts cs431-cotter Real-time process execution dispatch 39 Summary • Primary job of an OS is to schedule processes – Batch – Interactive – Real Time • Selection of scheduling algorithm depends on optimization criteria cs431-cotter 40 Questions • • • • • On a system with multilevel queue scheduling, would you expect to see the same scheduling algorithms used on all levels, or would you expect them to be different? Justify your answer. Consider the following set of processes, with the length of the CPU burst time given in milliseconds. The processes are assumed to have arrived in the order P1, P2, P3, P4. Process Job Size P1 10 P2 1 P3 2 P4 5 Of the scheduling algorithms FCFS, SJF, and RR(qt=2) which algorithm offers the best waiting time? In many multiprocessing systems, although there is a fixed limit to the amount of time that a job can keep control of the CPU, in practice, the jobs are released earlier. Why? The book talked about the multiple queueing system used by CTSS. If a process needed 30 quanta to complete, how many times would it be swapped in to complete? How does Lottery Scheduling work? What is its principle advantage over, say, Guaranteed Scheduling? cs431-cotter 41