Priority Scheduling

Notes

• This is chapter 5.3.4
• The book says that Shortest Job First is an instance of priority scheduling.
• A scheduler, looking at the jobs, decides which have the highest priorities.
• There is no fixed scale for high and low
  • Different OSs use different scales (0-7, 0 high, 1-4095, 4095 high)
  • They use low numbers for high priority.
• This matches the unix/linux nice command
  • Look at man page
  • This allows a user to be "nice" and lower the priority of their process
  • A range of -20 to 19
  • The normal user can only increase their nice level
  • renice 20 pid
  • The superuser can raise the priority level by assigning it a lower number.
• The book suggests a large number of ways to prioritize
  • time used
  • Resource requirements:
    • Memory used
    • Files open
    • burst history
  • But also external requirements set by users or admin
    • renice or nice
    • chrt (I don't know this well)
      • Set, change or check a process' priority
      • chrt -p pid
      • (as root) chrt -r number pid
        • Priorities between 1 (low) and 99 (high)
      • YOu can also set the queue (or scheduler)
      • Take a quick look
      • We will be back.
  • In a preemptive system, a process with a higher priority might interrupt the current process.
  • The system needs to take starvation into account
    • This can be countered with aging, or increasing a process' priority when it has not run
  • Frequently round robin is used when there are multiple processes with the same priority.
• Current systems implement multiple queues
  • This reduces the need to search all processes to schedule
  • Simply select a process from the highest level non-empty queue.
  • Real time, system, interactive, batch, idle
• Different scheduling algorithms can be applied to each of the queues
• To prevent starvation, systems can time-slice between queues
  • Real Time needs to execute so it has priority
  • The system level has 50% of the time
  • The interactive has 30% of the time
  • The batch queue has 20% of the time
• A queuing system, like interactive, might also have multiple queues
  • When a process enters it is put into a "fast" queue, say quantum 8
  • If it swaps out before the quantum expires it stays in that queue.
  • If it exceeds that time, it moves the the quantum 16 queue
  • This can be repeated until it reaches the FCFS with preemption queue
• look at man 7 sched for a basic overview of linux scheduling
  • EEVDF - Earliest Eligible Virtual Deadline First
  • Designed for multicore systems
  • Attempts to meet "virtual deadlines" based on how much time the process has used.
  • And distribute time evenly by keeping track how how much time a process is "owed".
  • This was added in 2024, based on a 1995 paper