CPU Scheduling

Objectives

1. Discuss the basics of cpu scheduling

Notes

• Keeping processes running on the CPU or CPUS or cores or ... is the fundamental operating system function.
• They change the description of a process slightly for these discussions
  • Most processes start out with a CPU Burst or a period of time when the process is active on the cpu.
  • This is followed by a I/O Burst or a period when the process is waiting for I/O.
  • This cycle repeats, see Figure 5.1 page 200.
• You can be sure that this has been studied in depth
  • On simulations
  • In the real world.
• The burst time of individual processes will vary but
  • Research shows that a pattern emerges in the long run.
  • See the figure on page 201.
  • Most processes fall into a small range of CPU bursts.
  • Very few have exceedingly long bursts.
  • This information is important when discussing process behavior WRT scheduling.
• Scheduling
  • This occurs when the/a processor(s) becomes idle (process blocked or times out)
  • The Scheduler selects the next task to run from the runtime queue/queues
  • We may have one, or multiple run time queues.
    • Different process priorities
    • Real time, normal and batch queue.
    • Or other situations.
  • We will definitely have different goals for the scheduler as well (discussed soon)
• In the queue, the process is represented by the Process Control Block (PCB)
  • We have discussed this.
  • In linux sched.h struct task_struct
  • All the information necessary to load the task into memory.
• A context switch then (114 -115 if you need to re-read it)
  • Is moving relevant information from the registers/memory/... to the PCB for the process that is being removed from the core
  • Loading the relevant information from the new process' PCB
  • Switching back to user mode
  • See figure 5.3 page 203
• Context switches are performed by the Dispatcher.
  • Since many of these occur, this needs to be efficient.
  • The time to perform context switch is called dispatcher latency.
  • grep ctxt /proc/pid/status will show the context switches for a process
    • Voluntary - switch because of resource request
    • involuntary - times out.
    • Take a look at reader and pi.
• Scheduling Terms
  • Scheduling is based on process switching
    1. from running to waiting due to service request
    2. from running to ready due to timeout interrupt
    3. from waiting to ready due to service finished
    4. from running to termination
  • They label 1 and 4 as cooperative or nonpreemptive
  • They label 2 and 3 as preemptive
  • In a nonpreemptive system, a process runs until it is terminated or needs a service.
    • And this can lead to poor performance.
  • Most modern os's are preemptive.
  • They state kernels can be preemptive or non-preemptive.
  • In an nonpreemeptive kernel, the kernel is allowed to complete tasks on behalf of the process without interruption.
    • But this leads to poor real-time performance.
    • So most modern kernels are preemptive