Part 3: Process Scheduling

• Basic Concepts
• Scheduling Criteria
• Scheduling Algorithms
– Uniprocessor (single-core)
scheduling
– Multiprocessor (multi-core)
scheduling

Operating Systems 3.1 Part 3 Process

 Basic Concepts
• CPU–I/O Bursts: Process execution
alternates between CPU executions and
I/O operations
– CPU burst: Duration of one CPU execution
cycle
– I/O burst: Duration of one I/O operation
(wait time)

• CPU scheduling objective: Keep the CPU

busy as much as possible (i.e., using
multiprogramming)

Operating Systems 3.2 Part 3 Process

 Alternating Sequence of CPU And I/O
Bursts

Operating Systems 3.3 Part 3 Process

 CPU Scheduler (Short-Term)
• Goal: Select one or more of the processes in
the ready queue, and allocate it to the CPU
(one process per core)

• Scheduler may run whenever a process

changes
4. state2. 5.
In which
case(s) does
CPU
scheduling
have to occur
in order to
3. keep the CPU
1. Answer:
cores busy?
Only 1 and 5

Operating Systems 3.4 Part 3 Process

 Types of CPU Schedulers

1.Nonpreemptive means that once the CPU has

been allocated to a process, the process keeps
the CPU until it voluntarily releases the CPU
either by terminating or requesting I/O/event
wait
– Scheduling is nonpreemptive if it happens
only for transitions 1 and 5 (also 4, when a
CPU core is idle)

2.Preemptive means that the CPU can be taken

away from a running process at any time by
the scheduler
– Scheduling is preemptive if it also happens
for transitions 2, 3.5
Operating Systems 3, 4 or at any other time
Part 3 Process
 Scheduling Objectives

System-wide Objectives:
1. Max. CPU Utilization: Keep the CPU as
busy as possible
– Percentage of time during which the
CPU cores are busy executing
processes.
– (Sum of execution time on each
core) divided by (total time
multiplied by number of cores)
2. Max. Throughput: Number of processes
that complete their execution per unit
Operating Systems 3.6 Part 3 Process
 Scheduling Objectives (Cont.)
Individual Process Objectives:
1. Min. Turnaround Time: Amount of time
to execute a particular process, i.e.,
from the time of creation to the time of
termination
– Time elapsed between “admitted”
and “exit” transitions for a process
– Average over all processes is
denoted as average turnaround time

Operating Systems 3.7 Part 3 Process

 Scheduling Objectives (Cont.)
2. Min. Waiting Time: Amount of time a
process has been waiting in the “ready”
state
– We do not consider time in “waiting”
state; why?
– Turnaround time components: CPU burst
(running), I/O burst (waiting) and waiting
time (ready)
– If all processes have a single CPU burst
(no I/O), then waiting time = turnaround
time – CPU burst

3. Min. Response Time:

Operating Systems 3.8 Time from a Part 3 Process
 Scheduling Algorithms: Uniprocessor
System

1. First-Come, First-Served (FCFS)

2. Shortest Job First (SJF)
3. Priority-based Scheduling
4. Round Robin (RR)
5. Multilevel Queue Scheduling
We assume a single CPU burst for each
process, a single CPU core and focus on
the average waiting time
Operating Systems 3.9 Part 3 Process
 First-Come, First-Served (FCFS)
Scheduling
• Example: Process CPU Burst Length
P1 24
P2 3
P3 3
• Suppose that processes arrive in the order P1, P2, P3, all
at time 0

• The Gantt Chart

P
for the schedule is:
P P3
1 2

0 24 27 30

• Waiting time for P1 = 0; P2 = 24; P3 = 27

• Average waiting time: 3.10
Operating Systems (0 + 24 + 27)/3 = 17 Part 3 Process
 FCFS Scheduling (Cont.)
• Suppose that the processes arrive in the
order P2, P3, P1, again all at time 0

• 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
– Much better than the previous case

Operating Systems 3.11 Part 3 Process

 FCFS Scheduling: Properties

• Is FCFS preemptive or nonpreemptive?

– Nonpreemptive, because processes
have to voluntarily release the CPU
once allocated

• Main problem with FCFS is the convoy

effect
– Short processes suffer increased
waiting times due to an earlier arrived
long process
• We saw this effect in the previous
Operating Systems 3.12 Part 3 Process
 Shortest-Job-First (SJF)
Scheduling
• Prioritize processes based on their CPU burst
lengths
– Shorter burst implies higher priority
– Intuitive way to handle the convoy effect of
FCFS

• Two schemes:
1. Nonpreemptive: Once core is given to a
process, it cannot be preempted until it
completes its CPU burst
2. Preemptive (also known as Shortest
Remaining-Time First or SRTF): If a newly
created process has CPU burst length less
than the remaining CPU burst of currently
Operating Systems 3.13 Part 3 Process
 Example of Nonpreemptive SJF
Process Arrival Time Burst Time
P1 0.0 7
P2 2.0 4
P3 4.0 1
P4 5.0 4

• SJF (Nonpreemptive) Gantt Chart:

P1 P3 P2 P4

0 3 7 8 12 16

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

Operating Systems 3.14 Part 3 Process

 Example of Preemptive SJF (SRTF)
Process Arrival Time Burst Time
P1 0.0 7 5 0
P2 2.0 4 2 0
P3 4.0 1
0
P4 5.0 4 0

• SJF (Preemptive) Gantt Chart:

P1 P2 P3 P2 P4 P1

0 2 4 5 7 11 16
P1 P2 P3 P4
• Average waiting time = (9 + 1 + 0 +2)/4 = 3

Operating Systems 3.15 Part 3 Process

 Priority-based Scheduling

• A priority number (integer) is associated with

each process

• CPU is allocated to the process with the highest

priority
– Usually, smallest integer implies highest
priority
– Two schemes: Preemptive and
nonpreemptive
– FCFS: Priority based on arrival order
– SJF: Priority based on CPU burst length

• Problem of Starvation: Lower priority processes

may never execute in a heavily loaded system
Operating Systems 3.16 Part 3 Process
 Round Robin (RR) Scheduling

• Fixed time quantum for scheduling (q): A

Process is allocated CPU for q time units,
preempted thereafter and inserted at the
end of the ready queue
– Processes are scheduled cyclically
based on q
– Usually q is small, around 10-100
milliseconds

• Performance:
– n processes in ready queue implies
waiting time is no more than (n-1)q
time units
Operating Systems 3.17 Part 3 Process
 Example: RR with Time Quantum
= 20
Process Burst Time
P1 53 33

P2 17 0

P3 68 48

P4 24 4

• All processes come at time 0 in the order of

P1, P2, P3, P4

• RR scheduling Gantt Chart:

P1 P2 P3 P4 P1 P3 P4 P1 P3 P3

0 20 37 57 77 97 117 121 134 154 162

• Typically, higher average waiting/turnaround

time than SJF, but better response time
Operating Systems 3.18 Part 3 Process
 Smaller Time Quantum Increases Context
Switches

Operating Systems 3.19 Part 3 Process

 Turnaround Time Varies With Time Quantum

• No well-
defined
relationship
between
average
Assume arrival turnaround
order is P1, P2, time and
P3, P4 all at quantum size
time 0
• Average
turnaround
time remains
fixed once
quantum size
is larger than
maximum CPU
burst length
Operating Systems 3.20 Part 3 Process
 Multilevel Queue Scheduling

• Different processes have different

requirements
– Foreground processes like those
handling I/O need to be interactive (RR
is preferred)
– Background processes need NOT be
interactive; scheduling overhead can
be low (FCFS)

• Solution: Multi-level Queue Scheduling

– Ready queue is partitioned into several
queues
Operating Systems 3.21 Part 3 Process
 Multilevel Queue Scheduling
(Cont.)
• Two schemes for inter-queue scheduling
– Fixed priority scheduling
• Queues served in priority order
(e.g., foreground before
background)
• Starvation for lower priority queues

– Time-slice based scheduling

• Each queue gets a fixed time
quantum on the CPU (e.g., 80ms for
foreground, 3.22
Operating Systems
20ms for background,
Part 3 Process
 Scheduling Algorithms: Multiprocessor
System

1. Partitioned Scheduling (Asymmetric

Multiprocessing - AMP)
2. Global Scheduling (Symmetric
Multiprocessing - SMP)

We assume a multi-core CPU and that

each process may only execute on one
CPU core at any time instant

Operating Systems 3.23 Part 3 Process

 Partitioned Scheduling (AMP)

• Processes are partitioned apriori (at

process creation time) among CPU
cores
– Each process is mapped to one core
– Separate uniprocessor CPU
• Advantages
scheduling for each core (asymmetric
– scheduling)
Core-specific scheduling strategy is
feasible (FCFS, SJF, RR, multi-level,
etc.)
– Easy and simple extension of
uniprocessor CPU scheduling to
multiprocessor
Operating Systems CPU scheduling
3.24 Part 3 Process
 Partitioned Scheduling: Issues

• How to map cores to processes?

– Poor mapping can lead to unequal
loads: a core is idling while another is
overloaded
• May reduce system/process
performance
• NP-Hard problem: Similar to the
well-known knapsack problem!
• Heuristics such as best-fit could be
used if CPU burst lengths are
known
Operating Systems 3.25 Part 3 Process
 Global Scheduling (SMP)

• One or more ready queues for the entire

system (no mapping of queues to CPU
cores)
– When any core becomes idle, CPU
scheduler runs and allocates the next
process to execute
– Process selection based on strategy
applied globally or symmetrically
across all cores (FCFS, SJF, etc.)
– It is possible that the same process
may execute on different cores at
different times
Operating Systems 3.26 Part 3 Process
 Example of SRTF (dual-core)
Process Arrival Time Burst Time
P1 0.0 7 3 0
P2 2.0 4 0
P3 4.0 1
0
P4 5.0 4 0
• SJF (Preemptive) on dual-core Gantt Chart:

Core P1 P3 P1
1
Core P2 P4
2
0 2 4 5 6 8 10
P1 P2 P3 P4

• Average waiting time = (1 + 0 + 0 + 1)/4 = 0.5

Operating Systems 3.27 Part 3 Process
 Example of RR with q = 20 (dual-
core)
Process Burst Time
P1 53 33 13 0
P2 17 0

P3 68 48 28 0

P4 24 4 0

• All processes come at time 0 in the order of

P1, P2, P3, P4

• RR0 scheduling
20 40 60 73
dual-core Gantt Chart:
P1 P4 P3 P1
Core
1
Core P2 P3 P1 P4 P3
2
0 17 37 57 61 89
Operating Systems 3.28 Part 3 Process
 Global Scheduling: Issues

• Implementation complexity is high when

compared to partitioned scheduling
– Need to synchronize clocks across
cores
– Global strategy for process selection
combining all the cores
• Implementation overhead is high
when compared to partitioned
scheduling
– Process could context switch from
one core and be scheduled on another
(private core-specific
Operating Systems 3.29 cache data Part 3 Process