 CPU Scheduling

Dr. Manmath N. Sahoo

Dept. of CSE, NIT Rourkela
CPU-I/O Burst Cycle
 CPU–I/O Burst Cycle – Process
execution consists of a cycle of CPU
execution and I/O wait.
 CPU burst is length of time process
needs to use CPU before it next
makes a I/O request.
 I/O burst is the length of time
process spends waiting for I/O to
complete.

Dr. Manmath N. Sahoo (CS) 2

Histogram of CPU-burst Times
Typical CPU burst distribution

Dr. Manmath N. Sahoo (CS) 3

CPU Scheduler

 Allocates CPU to one of processes that are ready

to execute (in ready queue)
 CPU scheduling decisions may take place when a
process:
1. Switches from running to waiting state (e.g. I/O request)
2. Terminates
3. Switches from waiting to ready (e.g. I/O completion)
4. Switches from running to ready state (e.g. priority based
interruption)

Dr. Manmath N. Sahoo (CS) 4

Dispatcher

 Dispatcher gives control of the CPU to the process

selected by the short-term scheduler; this involves:
• switching context
• switching to user mode (if CPU was executing the previous
process in kernel mode and preempted)
• jumping to the proper location in the user program to restart that
program (i.e., last action is to set program counter)
 Dispatch latency – time it takes for the dispatcher
to switch between processes and start new one
running

Dr. Manmath N. Sahoo (CS) 5

Scheduling Criteria

 CPU utilization: i.e. CPU usage – want to maximize

 Throughput: number of processes that complete their
execution per time unit – want to maximize
 Turnaround time: amount of time to execute a particular
process – want to minimize
 Waiting time: amount of time a process has been waiting
in the ready queue – want to minimize
 Response time: amount of time it takes from when a job
was submitted until it initiates its first response (output) –
want to minimize

Dr. Manmath N. Sahoo (CS) 6

First-Come, First-Served Scheduling

 Schedule: order of arrival of process in ready queue

Example: Process Burst Time
P1 24
P2 3
P3 3
 Suppose that the processes arrive in the order: P1 , P2 , P3. The Gantt
Chart for the schedule is:

P1 P2 P3

0 24 27 30
 Waiting time for P1 = 0; P2 = 24; P3 = 27
 Average waiting time: (0 + 24 + 27)/3 = 17

Dr. Manmath N. Sahoo (CS) 7

FCFS Scheduling

 Suppose that the processes arrive in the

order P2 , P3 , P1 .
 The Gantt chart for the schedule is:
P2 P3 P1

0 3 6 30

 Waiting time for P1 = 6; P2 = 0; P3 = 3

 Average waiting time: (6 + 0 + 3)/3 = 3

Dr. Manmath N. Sahoo (CS) 8

FCFS Scheduling

+ Simple to implement by FIFO queue.

+ Code is easy to write and understand
- Average waiting time is not minimal
- Not suitable for time-sharing systems
- Convoy effect: short I/O intensive processes
behind long CPU intensive process
 NOTE: FCFS Scheduling is non-preemptive
Dr. Manmath N. Sahoo (CS) 9
 Large CPU bound process
• C1 (1200ms): 1000ms computation + 200ms I/O
• 200cpu – 50io – 200cpu – 50io – 200cpu – 50io – 200cpu – 50io – 200cpu

 Small I/O bound processes

• I1, I2, …, I10
• (8.5ms): 2.5ms computation + 8ms I/O
• 0.5cpu – 2io – 0.5cpu – 2io – 0.5cpu – 2io – 0.5cpu – 2io – 0.5cpu

 May lead to less CPU utilization and I/O utilization →

Convoy effect

Dr. Manmath N. Sahoo (CS) 10

Shortest-Job-First (SJF) Scheduling

 Schedule: order of the CPU burst of the

processes.
 Two schemes:
• Non-Preemptive SJF / Shortest Job Next (SJN)
• once CPU given to the process it cannot be preempted until
completes its CPU burst.
• Preemptive SJF / Shortest Remaining Time First
(SRTF)
• if a new process arrives with CPU burst length less than
remaining time of current executing process, preempt.
Dr. Manmath N. Sahoo (CS) 11
Non-Preemptive SJF (SJN)

Process Arrival Time Burst Time

P1 0 7
P2 2 4
P3 4 1
P4 5 4
 Gantt Chart:
P1 P3 P2 P4

0 3 7 8 12 16

 Average waiting time = (0 + 6 + 3 + 7)/4 = 4

Dr. Manmath N. Sahoo (CS) 12
Preemptive SJF (SRTF)

Process Arrival Time Burst Time

P1 0 7
P2 2 4
P3 4 1
P4 5 4
 Gantt Chart:
P1 P2 P3 P2 P4 P1

0 2 4 5 7 11 16

 Average waiting time = (9 + 1 + 0 +2)/4 = 3

Dr. Manmath N. Sahoo (CS) 13

Example

Process Arrival Time Burst Time

P1 0 16
P2 2 8
P3 4 12
P4 6 2
P5 8 4

 Draw Gantt chat and find the average waiting time

in each of the following cases:
• SRTF Scheduling
• SJN Scheduling
• FCFS Scheduling

Dr. Manmath N. Sahoo (CS) 14

 Answer to Example Process Arrival Time Burst Time
P1 0 16
P2 2 8
P3 4 12
P4 6 2
P5 8 4
 SRTF

P1 P2 P2 P4 P2 P5 P3 P1
0 2 4 6 8 12 16 28 42

 Average waiting time = (26+6+12+0+0)/5

= 8.8

Dr. Manmath N. Sahoo (CS) 15

Answer to Example Process Arrival Time Burst Time
P1 0 16
P2 2 8
P3 4 12
P4 6 2
P5 8 4
 SJN
P1 P4 P5 P2 P3
0 16 18 22 30 42

 Average waiting time = (0+20+26+10+10)/5

= 13.2

Dr. Manmath N. Sahoo (CS) 16

Answer to Example Process Arrival Time Burst Time
P1 0 16
P2 2 8
P3 4 12
P4 6 2
P5 8 4
 FCFS
P1 P2 P3 P4 P5
0 16 24 36 38 42

 Average waiting time = (0+14+20+30+30)/5

= 18.8

Dr. Manmath N. Sahoo (CS) 17

Answer to Example

Scheduling Algorithm Average Waiting Time

FCFS 18.8
SJN 13.2
SRTF 8.8

Dr. Manmath N. Sahoo (CS) 18

Shortest Job First

+ SRTF is Optimal: Minimum Average waiting time.

- Longer processes may be starved.
- Difficult to know the CPU burst of processes before their
execution

Dr. Manmath N. Sahoo (CS) 19

Round Robin (RR) Scheduling

 Each process gets a time quantum, usually 10-

100 milliseconds.
 After this time has elapsed, the process is
preempted and added to the end of the ready
queue.
 If there are n processes in the ready queue and
the time quantum is q, then each process gets
1/n of the CPU time.
 No process waits more than (n-1)q time units.

Dr. Manmath N. Sahoo (CS) 20

Example: RR with q= 20

Process Burst Time

P1 53
P2 17
P3 68
P4 24
 The Gantt chart is:
P1 P2 P3 P4 P1 P3 P4 P1 P3 P3

0 20 37 57 77 97 117 121 134 154 162

 Typically, higher average turnaround than SJF, but

better response.
Dr. Manmath N. Sahoo (CS) 21
Round Robin (RR) Scheduling

 FCFS + Preemption = RR
 Performance
• q large ⇒ behaves similar to FCFS
• q small ⇒ more context switch overhead

Dr. Manmath N. Sahoo (CS) 22

Modified RR Scheduling

 Based on variable time quantum

 A preempted process, on its next turn will
use an increased time quantum
 RR Scheduling
• P1: 300ms, q=20ms, time per context switch=2ms
• #of preemptions due to this process = 14
• Preemption overhead = 28ms

Dr. Manmath N. Sahoo (CS) 23

Modified RR Scheduling

 Modified RR Scheduling – q is doubled in

next turn.
• P1: 300ms, q=20ms (initial)
• #of preemptions due to this process = 3.
(20+40+80+160=300)
• Preemption overhead = 6ms

Dr. Manmath N. Sahoo (CS) 24

Priority Scheduling

 A priority number (integer) is associated with each

process and the CPU is allocated to the process with
the highest priority
 Some schemes use smallest integer ≡ highest
priority.
 SJF is effectively priority scheduling where priority
is defined on next CPU burst time.
 FCFS is effectively priority scheduling where
priority is defined on arrival time.

Dr. Manmath N. Sahoo (CS) 25

Priority Scheduling

 Problem: Starvation – low priority

processes may never execute.
 Solution: Aging – as time progresses
increase the priority of the process.
• e.g. in every 10ms increase the priority of process by 1

Dr. Manmath N. Sahoo (CS) 26

Highest Response Ratio Next
(HRRN) Scheduling
 It gives priority to shorter jobs similar to SJN
 Implements Aging implicitly
 Non-preemptive dynamic priority based
algorithm with initial priority 1
 p=1+(wait time/burst time)

Dr. Manmath N. Sahoo (CS) 27

 Highest Response Ratio Next
(HRRN) Scheduling
Process Arrival Time Burst Time
P1 0 3
P2 2 6
P3 4 4
P4 6 5
P5 8 2

P1 P2 P3 P5 P4
0 3 9 13 15 20

Dr. Manmath N. Sahoo (CS) 28

Multi level Queue Scheduling

 Ready queue is partitioned into separate

queues.
 Each queue holds a particular category of
processes.
 Each queue can implement whatever
scheduling algorithm is most appropriate for
that type of processes.
Dr. Manmath N. Sahoo (CS) 29
Multi level Queue Scheduling

Dr. Manmath N. Sahoo (CS) 30

Multilevel Queue Scheduling

 Scheduling must be done between the

queues.
• Priority scheduling; But possibility of starvation.
• Time slice – each queue gets a certain amount of CPU
time which it can schedule amongst its processes;

Dr. Manmath N. Sahoo (CS) 31

Multilevel Queue Scheduling

 Objectives:
• To optimize turnaround time.
• To minimize response time (to make a system feel
responsive to interactive users).

THE CRUX:

HOW TO SCHEDULE WITHOUT PERFECT KNOWLEDGE?

How can we design a scheduler that both minimizes response time

for interactive jobs while also minimizing turnaround time without
a priori knowledge of job length and job behaviour?

Dr. Manmath N. Sahoo (CS) 32

Multilevel Feedback Queue Scheduling -
Basic Rules
 >1 number of queues, each assigned a different priority level
 Processes are executed from top priority queue to low priority
queue.

• Rule 1: If Priority(A) > Priority(B),

A runs (B doesn’t).
• Rule 2: If Priority(A) = Priority(B),
A & B run in RR/SJN/any other.

Dr. Manmath N. Sahoo (CS) 33

MLFQ: Basic Rules

 MLFQ varies the priority (dynamic) of a job

based on its observed behavior.
 If a process repeatedly relinquishes the CPU,
MLFQ will keep its priority high.
 If a job uses the CPU intensively for long
periods of time, MLFQ will reduce its priority.

Dr. Manmath N. Sahoo (CS) 34

MLFQ Priority Adjustment: Attempt#1
• Rule 3: When a job enters the system, it is placed at the highest
priority (the topmost queue).
• Rule 4a: If a job uses up an entire time slice while running, its
priority is reduced (i.e., it moves down one queue).
• Rule 4b: If a job gives up the CPU before the time slice is up, it
stays at the same priority level.

Dr. Manmath N. Sahoo (CS) 35

Example 1: A Single Long-Running Job

Dr. Manmath N. Sahoo (CS) 36

Example 2: Along Came A Short Job
(20ms) at 100ms

Dr. Manmath N. Sahoo (CS) 37

Example 3: What About I/O?
 An interactive job B needs the CPU only for 1ms before performing
an I/O of 9ms.

Dr. Manmath N. Sahoo (CS) 38

Problems With Current MLFQ

 Starvation to long jobs.

 A smart user could rewrite their program to
game the scheduler.
 A process may change its behavior over time
- from CPU bound to a phase of interactivity.

Dr. Manmath N. Sahoo (CS) 39

Attempt #2: The Priority Boost

 guarantees that CPU-bound jobs will make

some progress (even if it is not much).

• Rule 5: After some time period S, move all the jobs in the system to
the topmost queue.

Dr. Manmath N. Sahoo (CS) 40

Attempt #2: The Priority Boost
 interactive jobs need CPU only for 5ms before performing an I/O of 5ms.

(a) Without Priority Boost (b) With Priority Boost in every 50ms

Dr. Manmath N. Sahoo (CS) 41

Attempt #3: Better Accounting

 To prevent gaming of the scheduler, Rules 4a and 4b are

rewritten to the following single rule:

• Rule 4: Once a job uses up its time quantum at a given level

(regardless of how many times it has given up the CPU), its
priority is reduced (i.e., it moves down one queue).

Dr. Manmath N. Sahoo (CS) 42

Effect of anti-Gaming rule

(a) Without anti-Gaming rule (b) With anti-Gaming rule

Dr. Manmath N. Sahoo (CS) 43
Uniprocessor vs Multiprocessor

Single CPU with Cache Two CPUs with Caches

Dr. Manmath N. Sahoo (CS) 44

Multiprocessor Scheduling

 Single Queue Multiprocessor Scheduling

 Multi Queue Multiprocessor Scheduling

Dr. Manmath N. Sahoo (CS) 45

SQMS

 LTS loads all jobs to the common ready

queue.
 However, there are multiple STSs.
+ Simple to implement
- Synchronization overhead (lock overhead)
- Doesn’t preserve Cache affinity

Dr. Manmath N. Sahoo (CS) 46

SQMS

RR Scheduling on 04 CPUs:
processes bounce from CPU to CPU.
Solution – SQMS must adopt some affinity mechanism (requires migration)

Dr. Manmath N. Sahoo (CS) 47

SQMS

Schedule with migration

Dr. Manmath N. Sahoo (CS) 48

MQMS

 2 Processors: CPU0, CPU1.

• A goes to CPU0
• B goes to CPU1
• C goes to CPU0
• D goes to CPU1

Dr. Manmath N. Sahoo (CS) 49

MQMS
+ Synchronization overhead (lock overhead)
+ Preserves Cache affinity
- Load imbalance

Dr. Manmath N. Sahoo (CS) 50

MQMS
- Load imbalance
• C finishes

• A finishes

Dr. Manmath N. Sahoo (CS) 51

MQMS

 Solution to load imbalance -> migration

Solution: Simply move one of B or D to CPU 0

Dr. Manmath N. Sahoo (CS) 52

MQMS

A single migration doesn’t help.

Solution: In every 40ms migrate one job.

Dr. Manmath N. Sahoo (CS) 53