Process scheduling

Download Report

Transcript Process scheduling

Chapter 5: CPU Scheduling
Operating System Concepts – 8th Edition,
Silberschatz, Galvin and Gagne ©2009
Chapter 5: CPU Scheduling
 Basic Concepts
 Scheduling Criteria
 Scheduling Algorithms
 Thread Scheduling
 Multiple-Processor Scheduling
 Linux Example
Operating System Concepts – 8th Edition
5.2
Silberschatz, Galvin and Gagne ©2009
Objectives
 To introduce CPU scheduling, which is the basis for multiprogrammed
operating systems
 To describe various CPU-scheduling algorithms
 To discuss evaluation criteria for selecting a CPU-scheduling algorithm for a
particular system
Operating System Concepts – 8th Edition
5.3
Silberschatz, Galvin and Gagne ©2009
Basic Concepts
 The objective of multiprogramming is to have some process
running at all time, to Maximum CPU utilization.
 CPU–I/O Burst Cycle – Process execution consists of a cycle of
CPU execution and I/O wait
 Process execution begins with a CPU burst that is followed by an
I/O burst, which is followed by another CPU burst , then another I/O
burst , and so on,.. The final CPU burst ends the process.
 CPU burst distribution

large number of short CPU bursts and a small number of long
CPU bursts.
 An I/O –bound program has many short CPU bursts.
 A CPU –bound program has few long CPU bursts.
Operating System Concepts – 8th Edition
5.4
Silberschatz, Galvin and Gagne ©2009
Histogram of CPU-burst Times
Operating System Concepts – 8th Edition
5.5
Silberschatz, Galvin and Gagne ©2009
Alternating Sequence of CPU And I/O Bursts
Operating System Concepts – 8th Edition
5.6
Silberschatz, Galvin and Gagne ©2009
CPU Scheduler
 When the CPU becomes idle, the OS must Select from among the
processes in memory that are ready to execute, and allocates the CPU to
one of them.
 The selection process is carried out by the short-term scheduler (CPU
scheduler ).
 CPU scheduling decisions may take place when a process:
1. Switches from running state to the waiting state(result of I/o request or
wait for the termination of one of the child processes).
2. Switches from running state to ready state(interrupt).
3. Switches from waiting state to ready state(completion of I/O)
4. Terminates
 Scheduling under 1 and 4 is nonpreemptive or cooperative.
 All other scheduling is preemptive
Operating System Concepts – 8th Edition
5.7
Silberschatz, Galvin and Gagne ©2009
Preemptive scheduling
 Under nonpreemptive scheduling, once the CPU has been allocated to a
process, the process keeps the CPU until it releases the CPU either by
terminating or by switching to the waiting state.
 Windows 95 and all subsequent versions of windows OS have used
preemptive scheduling.
Operating System Concepts – 8th Edition
5.8
Silberschatz, Galvin and Gagne ©2009
Dispatcher
 The Dispatcher is the module that gives control of the CPU to the
process selected by the short-term scheduler; this involves:

switching context

switching to user mode

jumping to the proper location in the user program to restart
that program
 It should be fast.
 Dispatch latency – the time it takes for the dispatcher to stop one
process and start another running
Operating System Concepts – 8th Edition
5.9
Silberschatz, Galvin and Gagne ©2009
Scheduling Criteria
 CPU utilization – keep the CPU as busy as possible.
 Throughput – # of processes that complete their execution per
time unit(10 processes/second)
 Turnaround time – amount of time to execute a particular
process(the interval from the time of submission of a process to the
time of completion, waiting to get into memory, waiting in the ready
queue, exciting on the CPU, doing I/O).
 Waiting time – the amount of times a process has been waiting in
the ready queue
 Response time – amount of time it takes from when a request was
submitted until the first response is produced, not output (for timesharing environment)
Operating System Concepts – 8th Edition
5.10
Silberschatz, Galvin and Gagne ©2009
Scheduling Algorithm Optimization Criteria
 Max CPU utilization
 Max throughput
 Min turnaround time
 Min waiting time
 Min response time
Operating System Concepts – 8th Edition
5.11
Silberschatz, Galvin and Gagne ©2009
First-Come, First-Served (FCFS) Scheduling
 Jobs are scheduled in order of arrival
 When a process enters the ready queue, its PCB is linked onto the tail
of the queue.
 When the CPU is free, it is allocated to the process at the head of the
queue (the running process is then removed from the queue).
 Disadvantages:
 Non-preemptive : once the CPU is allocated to a process, the process
keeps the CPU until it releases it, either by terminating or requesting I/O.
 The average waiting time is often quite long.
Operating System Concepts – 8th Edition
5.12
Silberschatz, Galvin and Gagne ©2009
example
Process
P1
P2
Burst Time
24
3
P3
3
 Suppose that the processes arrive in the order: P1 , P2 , P3
The Gantt Chart for the schedule is:
P1
0
P2
24
P3
27
30
 Waiting time for P1 = 0; P2 = 24; P3 = 27
 Average waiting time: (0 + 24 + 27)/3 = 17
Operating System Concepts – 8th Edition
5.13
Silberschatz, Galvin and Gagne ©2009
FCFS Scheduling (Cont)
Suppose that the processes arrive in the order
P2 , P3 , P1
 The Gantt chart for the schedule is:
P2
0
P3
3
P1
6
30
 Waiting time for P1 = 6; P2 = 0; P3 = 3
 Average waiting time: (6 + 0 + 3)/3 = 3
 Much better than previous case
 Convoy effect as short processes go behind long process lower
CPU and device utilization.
Operating System Concepts – 8th Edition
5.14
Silberschatz, Galvin and Gagne ©2009
Shortest-Job-First (SJF) Scheduling
 This algorithm Associates with each process the length
of its next CPU burst. Use these lengths to schedule the
process with the shortest time, if the next CPU bursts of
two processes are the same, FCFS scheduling is used.
 Two schemes:

Nonpreemptive – once CPU given to the process it
cannot be preempted until completes its CPU burst

Preemptive – if a new process arrives with CPU burst
length less than remaining time of current executing
process, preempt. This scheme is known as the
Shortest-Remaining-Time-First (SRTF)
Operating System Concepts – 8th Edition
5.15
Silberschatz, Galvin and Gagne ©2009
Examples of SJF
Example1:
Process
Arrival Time
Burst Time
P1
0.0
6
P2
2.0
8
P3
4.0
7
P4
5.0
3
 SJF scheduling chart
P4
0
P3
P1
3
9
P2
16
24
 Average waiting time = (3 + 16 + 9 + 0) / 4 = 7

Compare with FCFS AWT=(0+6+14+21)/4=10.25
Operating System Concepts – 8th Edition
5.16
Silberschatz, Galvin and Gagne ©2009
Shortest-Job-First (SJF) Scheduling
Example2:
Process
P1
P2
P3
P4
 Non preemptive SJF
Arrival Time Burst Time
0
7
2
4
4
1
5
4
Average waiting time = (0 + 6 + 3 + 7)/4 = 4
P1
0
P3
4 5
2
7
P2
8
P4
12
P1(7)
16
P1‘s wating time = 0
P2‘s wating time = 6
P2(4)
P3‘s wating time = 3
P3(1)
P4‘s wating time = 7
P4(4)
Operating System Concepts – 8th Edition
5.17
Silberschatz, Galvin and Gagne ©2009
Shortest-Job-First (SJF) Scheduling
Example3:
Process Arrival Time Burst Time
P1
0
7
P2
2
4
P3
4
1
P4
5
4
 Preemptive SJF(SRTF)
P1(7)
P1
P2 P3
0
2
P1(5)
4
Average waiting time = (9 + 1 + 0 +2)/4 = 3
P2
5
P4
11
7
16
P1‘s wating time = 9
P2‘s wating time = 1
P2(4) P2(2)
P3‘s wating time = 0
P3(1)
P4‘s wating time = 2
P4(4)
Operating System Concepts – 8th Edition
P1
5.18
Silberschatz, Galvin and Gagne ©2009
Shortest-Job-First (SJF) Scheduling
 SJF is optimal – gives minimum average waiting
time for a given set of processes

The difficulty is knowing the length of the next CPU
request.
Operating System Concepts – 8th Edition
5.19
Silberschatz, Galvin and Gagne ©2009
Prediction of the Length of the Next CPU Burst
Operating System Concepts – 8th Edition
5.20
Silberschatz, Galvin and Gagne ©2009
Priority Scheduling
 A priority number (integer) is associated with each process
 The CPU is allocated to the process with the highest priority (smallest
integer  highest priority in Unix but lowest in Java)
 Equal-priority processes are scheduled in FCFS order.

Preemptive: preempt the CPU if the priority of the newly arrived process
is higher than the priority of the currently running process.

Nonpreemptive : put the new process at the head of the ready queue.
 SJF is a priority scheduling where priority is the predicted next CPU burst
time
 Problem  Starvation – low priority processes may never execute
 Solution  Aging – as time progresses increase the priority of the process
(for example : 1 every 15 minutes)
Operating System Concepts – 8th Edition
5.21
Silberschatz, Galvin and Gagne ©2009
Priority Scheduling
Example :
Process
P1
P2
P3
P4
p5
Burst Time priority
10
3
1
1
2
1
5
4
5
2
All arrived at time 0.
The Gantt chart for the schedule is:
P2
0
P1
P5
1
6
P3
16
18
P4
19
The AWT is (6 +0+ 16+18+1)/5 = 8.2
Operating System Concepts – 8th Edition
5.22
Silberschatz, Galvin and Gagne ©2009
Priority Scheduling
Example:
Process
arrival time Burst length Priority
P1
0
10
3
P2
0
1
1
P3
0
2
4
P4
0
1
5
P5
3
5
2
 Gantt chart: Non-preemptive priority scheduling
P2 P1
0
P5
1
11
P3
P4
16
18
19
 Gantt chart: Preemptive priority scheduling
P2
0
P1
1
P5
P1
3
8
Operating System Concepts – 8th Edition
P3
16
5.23
P4
18
19
Silberschatz, Galvin and Gagne ©2009
Round Robin (RR)
 Is designed especially for time-sharing systems.
 Similar to FCFS, but it is Preemptive to enable
the system to switch between processes.
 Each process gets a small unit of CPU time (time
quantum or time slice), usually 10-100
milliseconds.
 The Ready queue is FIFO (new processes are
added to the tail of the queue.)
 The CPU scheduler picks the first process from
the ready queue ,set a timer to interrupt after 1
time quantum, and dispatch the process.
Operating System Concepts – 8th Edition
5.24
Silberschatz, Galvin and Gagne ©2009
Round Robin (RR)
 One of two things will happen

The process may have a CPU burst of < 1 time
quantum the process itself will release the CPU
voluntarily.

The CPU burst of the currently running process > 1 time
quantum  the timer will go off and will cause an
interrupt to the OS.  a context switch will be executed,
and the process will be put at the tail of the ready queue.
 The CPU scheduler will then select the next process in the
ready queue.
 Typically, higher average turnaround than SJF, but better
response
Operating System Concepts – 8th Edition
5.25
Silberschatz, Galvin and Gagne ©2009
Round Robin (RR)
Example1:
Time quantum = 4
Process
P1
P2
P3
 The Gantt chart is:
P1
P2
P3
Burst Time
24
3
3
P1
0
10
14
4
7
 AWT(6(10-4)+4+7)/3 = 5.66
Operating System Concepts – 8th Edition
P1
P1
18 22
5.26
P1
26
P1
30
Silberschatz, Galvin and Gagne ©2009
Round Robin (RR)
 Example2:
 Time quantum = 20
Process
P1
P2
P3
P4
Burst Time
53
17
68
24
Wait Time
57 +24 = 81
20
37 + 40 + 17= 94
57 + 40 = 97
P1 P2 P3 P4 P1 P3 P4 P1 P3 P3
0 20 37 57 77 97 117 121 134 154 162
P1(33)
24P1(13)
P1(53)
57
P2(17) 20
P3(68)
P4(24)
37
57
P3(48)
40
40
17
P3(28) P3(8)
P4(4)
Average wait time = (81+20+94+97)/4 = 73
Operating System Concepts – 8th Edition
5.27
Silberschatz, Galvin and Gagne ©2009
Round Robin (RR)
 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 in chunks of at most q time units at once. No
process waits more than (n-1)*q time units until its next
time quantum. (Ex: 5 processes, TQ = 20 milliseconds,
each process will get up to 20 milliseconds every 100
milliseconds.
 The Performance of RR depends heavily on the size of
the TQ.

TQ large  FCFS

TQsmall  TQ must be large (but not too large)with
respect to context switch time, otherwise overhead is
too high
Operating System Concepts – 8th Edition
5.28
Silberschatz, Galvin and Gagne ©2009
Time Quantum and Context Switch Time
Operating System Concepts – 8th Edition
5.29
Silberschatz, Galvin and Gagne ©2009
Turnaround Time Varies With The Time Quantum
The average TurnAroundTime of a
set of process does not necessarily
improve as the TQ size increase.
The AVG TAT can be improved if
most process finish their next CPU
burst in a single time quantum.
Operating System Concepts – 8th Edition
5.30
Silberschatz, Galvin and Gagne ©2009
Multilevel Queue
 Processes are classified into different groups.
 Each group have different response-time requirements different scheduling
needs.
 A multilevel queue scheduling algorithm partitions the Ready queue into
separate queues:
foreground (interactive)
background (batch)
 Each queue has its own scheduling algorithm

foreground – RR

background – FCFS
 Scheduling must be done between the queues

Fixed priority preemptive scheduling; (i.e., serve all from foreground then
from background). Possibility of starvation.

Time slice – each queue gets a certain amount of CPU time which it can
schedule amongst its processes; i.e., 80% to foreground in RR, 20% to
background in FCFS
Operating System Concepts – 8th Edition
5.31
Silberschatz, Galvin and Gagne ©2009
Multilevel Queue Scheduling
Operating System Concepts – 8th Edition
5.32
Silberschatz, Galvin and Gagne ©2009
Multilevel Feedback Queue
 Implement multiple ready queues

Different queues may be scheduled using different algorithms
 Just like multilevel queue scheduling, but assignments are not
static
 Multilevel feedback queue-scheduling algorithm allows a
process to move between the various queues; aging can be
implemented this way
 Multilevel-feedback-queue scheduler defined by the
following parameters:

number of queues

scheduling algorithms for each queue

method used to determine when to upgrade and downgrade a
process
 The most general CPU-scheduling algorithm.
 The most complex algorithm.
Operating System Concepts – 8th Edition
5.33
Silberschatz, Galvin and Gagne ©2009
Example of Multilevel Feedback Queue
 Three queues:

Q0 – RR with time quantum 8 milliseconds

Q1 – RR time quantum 16 milliseconds

Q2 – FCFS
 Scheduling

A new job enters queue Q0 which is served FCFS. When it gains CPU,
job receives 8 milliseconds. If it does not finish in 8 milliseconds, job is
moved to queue Q1.

At Q1 job is again served FCFS and receives 16 additional milliseconds.
If it still does not complete, it is preempted and moved to queue Q2.

AT Q2 job is served FCFS only when queue 0 and queue 1 are empty.
Operating System Concepts – 8th Edition
5.34
Silberschatz, Galvin and Gagne ©2009
Multilevel Feedback Queues
Operating System Concepts – 8th Edition
5.35
Silberschatz, Galvin and Gagne ©2009
Thread Scheduling
 Distinction between user-level and kernel-level threads
 On OSs that support them, it is the kernel-level threads-
not processes- that are being scheduled by OS.
 User-level threads are managed by a thread library and
the kernel is unaware of them.
 To run on CPU, the user level threads must be mapped to
an associated kernel-level thread. It may use a lightweight
process(LWP).
contention scope:
 one distinction between user-level and kernel-level
threads lies in how they are scheduled.
Operating System Concepts – 8th Edition
5.36
Silberschatz, Galvin and Gagne ©2009
Thread Scheduling
 Many-to-one and many-to-many models, thread library schedules
user-level threads to run on LWP. Known as process-contention
scope (PCS) since scheduling competition takes place among
threads belonging to the same process.
 PCS is done according to preempt priority.

PTHREAD SCOPE PROCESS schedules threads using PCS
scheduling.
 Kernel thread scheduled onto available CPU is system-contention
scope (SCS) – competition takes place among all threads in system
 Systems using the one-to-one model schedule threads using only
SCS.

PTHREAD SCOPE SYSTEM schedules threads using SCS
scheduling.
Operating System Concepts – 8th Edition
5.37
Silberschatz, Galvin and Gagne ©2009
Multiple-Processor Scheduling
 CPU scheduling more complex when multiple CPUs are available
 Different rules for homogeneous processors (Identical processors in terms of
their functionality) or heterogeneous processors.
 Asymmetric multiprocessing:

All scheduling decisions, I/O processing, and other system activities handled by
a single processor – the master server.
 The other processors execute only user code.
 Simple because only one processor accesses the system data structures,
reducing the need for data sharing.
 Symmetric multiprocessing (SMP):

each processor is self-scheduling, all processes in common ready queue, or
each has its own private queue of ready processes
 Multiple processors try to access and update a common data structures. So,
scheduler must be programmed carefully.
 Must ensure that 2 processors don’t choose the same process.
Operating System Concepts – 8th Edition
5.38
Silberschatz, Galvin and Gagne ©2009
Linux Scheduling
 Linux Scheduler is a preemptive, priority-based algorithm
with 2 separate priority ranges:



Two priority ranges: time-sharing and real-time
A real-time range from 0 to 99 Longer time quantum
A nice value ranging from 100 to 140  Shorter time quantum
Operating System Concepts – 8th Edition
5.39
Silberschatz, Galvin and Gagne ©2009
Linux Scheduling
 The kernel maintains a list of all runnable tasks in a
runqueue data structure.
 Each runqueue contains two priority arrays :
 Active :
contains all tasks with time remaining in their time
slices
 Expired :
contains all expired tasks.
Operating System Concepts – 8th Edition
5.40
Silberschatz, Galvin and Gagne ©2009
List of Tasks Indexed According to Priorities
 The scheduler chooses the task with the highest priority
from the active array for execution on the CPU.
 When the active array is empty  the 2 arrays are
exchanged (the expired array becomes the active array,
and vice versa).
Operating System Concepts – 8th Edition
5.41
Silberschatz, Galvin and Gagne ©2009
Algorithm Evaluation
 Deterministic modeling
 takes a particular predetermined workload and defines the
performance of each algorithm for that workload
 More examples P: 214
 Simple and fast
 Requires exact numbers for input and its answers apply only
for those data
 Queueing models
 rate at which new processes arrive, ratio of CPU bursts to I/O
times, distribution of CPU burst times and I/O burst times can
be measured and then approximated or estimated
 result is a mathematical formula describing it
 From these it is possible to compute the average throughput,
utilization, waiting time, and so on
 difficult to work
Operating System Concepts – 8th Edition
5.42
Silberschatz, Galvin and Gagne ©2009
Algorithm Evaluation
 Simulations

run computer simulations of the different proposed algorithms
 data to drive the simulation can be randomly generated

better alternative when possible is to generate trace tapes
 expensive
 Implementation

The only completely accurate way to evaluate a scheduling algorithm is
to code it up, put it in the operating system, and see how it works.

high cost (coding and user reaction)
Operating System Concepts – 8th Edition
5.43
Silberschatz, Galvin and Gagne ©2009
Simulations
Operating System Concepts – 8th Edition
5.44
Silberschatz, Galvin and Gagne ©2009
Conclusion

We’ve looked at a number of different
scheduling algorithms.

Which one works the best is application
dependent.

General purpose OS will use priority based, round
robin, preemptive

Real Time OS will use priority, no preemption.
Operating System Concepts – 8th Edition
5.45
Silberschatz, Galvin and Gagne ©2009
End of Chapter 5
Operating System Concepts – 8th Edition,
Silberschatz, Galvin and Gagne ©2009