Lecture: Scheduling

• Problem: choosing which thread to run on the CPU next
• More generally: multiplex tasks onto limited resources
  • Cloud service, limited machines, lots of requests
  • Supermarket, limited cashiers, lots of customers

Definitions/Metrics

• Task: user request (mouse click, web request, shell command)
• Latency/Response time: user-perceived time to do a task
• Throughput: # of tasks done per unit of time
• Scheduling Overhead: time to switch from one task to another
• Fairness: how equal is the resource allocation
• Predictability: how consistent is the response time
• Starvation: lack of progress for one task (due to higher priority tasks)

Scheduling Mechanism

• Context switch
• Timer interrupt: preemption

Scheduling Policies

• First In First Out (FIFO)
  • Complete one task at a time, in the order they come in
  • Pros/cons?
  • Average response time: best case? worst case?
• Shortest Job First (SJF) / Shortest Remaining Time First (SRTF)
  • Schedule the task with the shortest remaining work to do
  • If a shorter task arrives, preempts the current task and switches to the new task
  • Pros/cons?
  • Average response time: best case? worst case?
• Round Robin (RR)
  • FIFO but each task only gets a fixed amount of time (time slice/quantum)
  • How to pick the time quantum?
  • Pros/cons?
  • Average response time: best case? worst case?
  • I/O bound vs CPU bound tasks
  • I/O bound: task that relies on/bottlenecked by I/O, uses little CPU and yields while waiting for I/O
  • CPU bound: task that relies on/bottlenecked by CPU, heavy on computation
• Multilevel Feedback Queue (MLFQ)
  • Goal: good average response time, but also want to be interactive
  • Multiple levels of RR queues, with increasing time quantum
  • Task starts at the top queue with smallest quantum
  • If task doesn't take up the whole time quantum, it moves up a queue (stays in the same queue if already at the top).
  • Otherwise it moves down a queue, as more CPU times are needed
  • Scheduler selects a task to run from the top queue in a RR fashion and works its way down
  • Periodically, all tasks are moved to the top queue (priority boost)
• Completely Fair Scheduler (CFS)
  • Ideal world: 2 processes each get 50% of CPU time, if one gets less, it should have priority to catch up to its fair time
  • Each CPU has a time-ordered red-black tree (ordered, self-balancing binary tree)
  • Time = time the task had been using the CPU
  • Always picks the task with least amount of time executed (to help it catch up)
  • As the task gains more CPU time, another tasks becomes the task with least amount of CPU time
• xk scheduler
  • How often is the scheduler invoked? kernel/trap.c
  • What scheduling policy does it have? kernel/proc.c