Scheduling Introduction

Notes

• This is chapter 5 of the book.
• Scheduling is selecting the next task to be given access to the CPU
  • We will at first assume, for simplicity, that there is only one cpu/core
  • But extend this later.
• We need to be aware of two extremes for processes
  • I/O intensive: tend to run in short bursts and need more data
  • Computationally intensive: tend to run in long bursts and need less data.
• Most processes are I/O intensive (figure 5.2 in the book)
• There are others in between, but we tend to think about these two
  • The I/O intensive processes tend to self-regulate
  • The computationally intensive processes are a problem for schedulers
• We tend to refer to the ready queue
  • This may indeed be a fifo
  • But it could be a ordered set of fifos
  • Or a tree
  • Or something else
  • Including just a linked list.
• The book lists four reasons scheduling must occur
  1. The current process switches from running to waiting (sleep, I/O request, ...)
  2. The current process is interrupted but is still able to run
  3. Another process switches from waiting to ready
  4. The current process terminates.
• When scheduling takes place for reasons 1 or 4, it is nonpreemptive
• When scheduling takes place for reasons 2 or 3 it is preemptive
• As we have seen, preemptive scheduling can cause concurrency issues.
• The dispatcher is normally the part of the CPU that does the context switch
  • Switches to kernel mode
  • Saves the state of the current process
  • Will possibly be required to do some memory management
  • Restores/establishes the state of the new process
  • Switches to user mode
• The dispatcher requires time, so switching between processes has an associated expense.
• Just for a clue
  • grep /ctx /proc/stat for context switches since boot
  • grep ctxt /proc/self/status for a process, substitute self with pid for another process.
• Generally, we have several goals a scheduler should or must achieve
  • CPU Utilization: keep the cpu as busy as possible
  • Throughput: Complete as many processes as possible
  • Turnaround time: complete processes as quickly as possible
  • Waiting time: Minimize the amount of time a process spends waiting
  • Response time: especially in interactive systems
• Note that some of these are contradictory and some are (semi) equivalent.