Lec 3-Scheduling(2)

Priority Schedulers

Simple Priority Scheduler

SJF, STCF are both priority schedulers

• Priority is CPU burst time

Also, FCFS and RR are priority schedulers

• FCFS's priority is arrival time
• RR's priority is dynamical

---

Problem with priority scheduling

• Starvation
  • High priority tasks can dominate the CPU
    $\Rightarrow$ low priority tasks will be starved

Possible solution

• Dynamically vary priorities based on proces behavior

---

Simple Priority Scheduler

• Schedule high priority tasks first
• No preemption
  
• Cannot automatically balance response and turnaround time

---

EDF: Earilest Deadline First

Each process has a deadline it must finish

• Tighter deadlines are given higher priority

• EDF is optimal (assuming preemption)
• But only useful if processes have known deadlines (e.g. Real-Time OS)

---

MLQ: Multilevel Queue

Key idea: Divide the ready queue in two

1. High priority queue for interactive processes
   • RR scheduling
2. Low priority queue for CPU bound processes
   • FCFS scheduling

Each process is assigned a priority on startup
Each queue is given a fixed amount of CPU time

---

Problems with MLQ

• Assume you can classify processes into high and low priority
  • How actually do this at run time
  • What if process's behavior changes over time
• High biased use of CPU time
  • How can we set the percentage
  • Convoy problems for low priority tasks

---

MLFQ: Multilevel Feedback Queue

Goals

• Minimize resopnse and turnaround time
• Dynamically adjust process priorities over time
  = Move processes between queue based on observed behavior
  • No prior knowledge about burst times or process behavior

First 4 Rules of MLFQ

1. If Priority(A) > Priority(B), A runs

2. If Priority(A) = Priority(B), A & B run in RR

3. Processes start at the highest priority
   = Assume initial process is interactive

4. Priority change rule
   4a) If a process uses an entire time slice, priority is reduced
   Each level queue has different time slice

   4b) If a process gives up the CPU before its time slice is up, priority level remains
   e.g.) Process performs I/O operations so that context switching occurs

1, 2. Process has time limit $\rightarrow$ priority is reduced
2. MLFQ $\approx$ STCF
3. Blocked on I/O $\rightarrow$ priority level remains

---

Problems with MLFQ

• By Rule 4b) Cheating can occur

---

MLFQ Rule 5: Priority Boost

• After period $S$, move all processes to the highest priority queue (=initialize)
• Solves two problems
  • Starvation
    • Low priority processes will eventually become high priority
  • Dynamic behavior
    • CPU bound process that has become interactive will be high priority

---

Revised Rule 4 (Cheat prevention)

Problem :

• Cheating (Repeatedly yield just before the time limit expires)

Solution :

• Once a process uses up its time allotment at a given priority, demote its priority
• Keep track of total CPU time during each time interval $S$
  • Instead of looking at continuous CPU time

---

MLFQ Rule Review

1. If Priority(A) > Priority(B), A runs
2. If Priority(A) = Priority(B), A & B run in RR
3. Processes start at the highest priority
   = Assume initial process is interactive
4. Once a process uses up its time allotment at given priority, demote it
5. After some time period $S$, move all processes to the highest prioirty queue

---

Note

• Achieve goals
  • MLFQ balances response time and turnaround time
  • MLFQ does not require prior knowledge about processes
• Parameters to tune
  • Number of queue
  • Distribution of CPU time
  • For each queue
    • Which scheduling regime
    • Time slice
• In practice
  • High priority queues - short time slices
  • Low priority queues - long time slices
    • Low priority - CPU bound - Longer time slice
  • Priority 0 sometimes reserved for OS processes
• Giving advice
  • Some OS allow users/processes to give the scheduler hints about priority
  • nice command on Linux
    • Run the command with adjusted priority
    • Values -20 ~ 19 is added to process priority