Scheduling

Overview

• scheduling: a policy that decides who should run on the CPU next
  • kernel schedules processes/threads onto the CPU
  • a process may perform a number of tasks/jobs (mouse click, web request, shell command)
• metrics for evaluting scheduling policies
• latency / turnaround time
  • user-perceived time to finish a task (including time waiting)
• throughput
  • the rate of task completion (# of tasks done per period of time)
• scheduling overhead
  • time it takes to perform scheduling (execution of scheduling policy + context switch)
• fairness
  • are tasks given similar times on the CPU & wait for similar amount of time
• starvation
  • lack of progress for one task due to higher priority tasks

Scheduling Policies

• first in first out (FIFO)
  • run each task to completion (or until it yields the CPU) in FIFO order
  • no preemption! minimal scheduling overhead
  • simple but high variance in latency based on the arrival ordering
  • what happens when a task blocks? what happens when it unblocks?
  • how well is FIFO according to the metrics?
• preemptive shortest job first (PSJF)
  • schedule the shortest job first (assumes that we know how long each task needs the CPU)
  • if a new shorter task arrives, preempts the current task and runs the new task
  • what happens when a task blocks? what happens when it unblocks?
  • how well is PSJF according to the metrics?
• round robin (RR)
  • FIFO but preemptive: each task gets x amt of time before it's preempted
  • what happens when the time quantum is too large? too small?
  • what happens when a task blocks? what happens when it unblocks?
  • how well is RR according to the metrics?
  • no starvation but unfair wait time for I/O bound tasks compared to CPU bound tasks
  • I/O bound: performance dependent on I/O
  • runs 1ms on the CPU, blocks 10ms for I/O
  • I/O bound jobs often block before the time slice expires (less time on the CPU)
  • but have to wait for CPU bound tasks who use the full time slice (long wait time)
  • CPU bound: performance dependent on CPU
  • runs 60ms on the CPU
  • how might we fix this?
  • prioritize tasks who spent less time on the CPU (CFS)
  • make I/O bound tasks wait less (MLFQ)
• multilevel feedback queue (MLFQ)
  • multiple RR queues with different time quantums
  • queues with shorter quantum has higher scheduling priority
  • tasks within a queue are schedueld in RR fashion
  • scheduler starts at the highest priority queue and moves down as higher priroty queues become empty
  • current task might be preempted if a higher priority queue has new tasks
  • tasks from lower queues are periodically moved up to the top priority queue to avoid starvation
  • how are tasks distributed to the queues?
  • a task starts at the top queue (shortest quantum, highest scheduling priority)
  • if a task uses up the quantum, it moves down a queue (longer quantum), as it needs more CPU
  • otherwise it stays in the same queue or moves up a queue