TDDI04
Concurrent Programming,
Operating Systems,
and Real-time Operating Systems
Overview: CPU Scheduling
§ CPU bursts and I/O bursts
§ Scheduling Criteria
§ Scheduling Algorithms
§ Multiprocessor Scheduling
CPU Scheduling
§ (Real-Time Scheduling
Lecture on real-time OS)
§ Scheduling Algorithm Evaluation
[SGG7] Chapter 5
Copyright Notice: The lecture notes are mainly based on Silberschatz’s, Galvin’s and Gagne’s book (“Operating System
Concepts”, 7th ed., Wiley, 2005). No part of the lecture notes may be reproduced in any form, due to the copyrights
reserved by Wiley. These lecture notes should only be used for internal teaching purposes at the Linköping University.
Acknowledgment: The lecture notes are originally compiled by C. Kessler, IDA.
Klas Arvidsson, IDA,
Linköpings universitet.
Basic Concepts
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.2
© Silberschatz, Galvin and Gagne 2005
CPU Scheduler
§ Maximum CPU utilization obtained with multiprogramming
§ CPU–I/O Burst Cycle –
Process execution consists of
a cycle of alternating
CPU execution and I/O wait
§ Selects from among the processes in memory that are ready
to execute, and allocates the CPU to one of them
§ CPU scheduling decisions may take place when a process:
1.
2.
3.
4.
5.
§ CPU burst distribution histogram
Becomes ready to run
Switches from waiting to ready
Switches from running to ready state
Switches from running to waiting state
Terminates
§ Scheduling under 1 to 3 is preemptive
•
The OS (may) choose to activate scheduling
§ Scheduling under 4 to 5 is nonpreemptive
•
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.3
© Silberschatz, Galvin and Gagne 2005
Dispatcher
switching context
6.4
© Silberschatz, Galvin and Gagne 2005
§ CPU utilization –
Keep the CPU as busy as possible
§ Throughput –
# of processes that complete their execution per time unit
switching to user mode
§ Turnaround time –
Total amount of time to complete execution of a particular process
jumping to the proper location in the user program
to restart that program
§ Waiting time –
Amount of time a process has been waiting in the ready queue
§ Dispatch latency
•
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
Scheduling Criteria
§ Dispatcher module gives control of the CPU to the process
selected by the short-term scheduler; this involves:
•
•
•
The process choose to activate OS scheduling (leave the CPU)
time it takes for the dispatcher to stop one process
and start another running
§ Response time –
Amount of time it takes from when a request was submitted
until the first response is produced,
not including output (for time-sharing environment)
§ Deadlines met? –
In real-time systems (later)
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.5
© Silberschatz, Galvin and Gagne 2005
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.6
© Silberschatz, Galvin and Gagne 2005
1
First-Come, First-Served (FCFS, FIFO)
Scheduling
Optimization Criteria
Process
P1
P2
P3
§ Max CPU utilization
§ Max throughput
§ Min turnaround time
Burst Time
24
3
3
§ Suppose that the processes arrive in the order: P1 , P2 , P3
§ The Gantt Chart for the schedule is:
§ Min waiting time
§ Min response time
P1
P2
0
24
P3
27
§ Waiting time for P1 = 0; P2 = 24; P3 = 27
FCFS normally used for
non-preemptive batch
scheduling, e.g. printer
queues (i.e., burst time
= job size)
Waiting time Pi = start time Pi – time of arrival for Pi
§ Average waiting time: (0 + 24 + 27) / 3 = 17
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
© Silberschatz, Galvin and Gagne 2005
6.7
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
30
© Silberschatz, Galvin and Gagne 2005
6.8
FCFS Scheduling (Cont.)
Shortest-Job-First (SJF) Scheduling
Suppose that the processes arrive in the order
P2 , P3 , P1
§ Associate with each process the length of its next CPU burst.
§ The Gantt chart for the schedule is:
§ Two schemes:
P2
0
P3
3
§ Use these lengths to schedule the shortest ready process
P1
6
30
§ Waiting time for P1 = 6; P2 = 0, P3 = 3
§ Average waiting time: (6 + 0 + 3)/3 = 3
•
nonpreemptive SJF – once CPU given to the process, it
cannot be preempted until it completes its CPU burst
•
preemptive SJF – preempt if a new process arrives with
CPU burst length less than remaining time of current
executing process.
- much better!
> Also known as Shortest-Remaining-Time-First (SRTF)
§ SJF is optimal –
§ Convoy effect – short process behind long process
• Idea: shortest job first?
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
© Silberschatz, Galvin and Gagne 2005
6.9
Example of Non-Preemptive SJF
Process Arrival Time
P1
0.0
P2
P3
P4
4
1
4
P2
P3
P4
P3
7
P2
8
P4
12
P1
16
§ Average waiting time = (0 + 6 + 3 + 7) / 4 = 4
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
© Silberschatz, Galvin and Gagne 2005
6.10
Burst Time
7
2.0
4.0
5.0
4
1
4
§ with preemptive SJF:
P1
3
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
Process Arrival Time
P1
0.0
§ with non-preemptive SJF:
0
gives minimum average waiting time
for a given set of processes
Example of Preemptive SJF
Burst Time
7
2.0
4.0
5.0
•
6.11
© Silberschatz, Galvin and Gagne 2005
0
P2
2
P3
4
P2
5
P4
P1
11
7
16
§ Average waiting time = (9 + 1 + 0 +2) / 4 = 3
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.12
© Silberschatz, Galvin and Gagne 2005
2
Predicting Length of Next CPU Burst
Examples of Exponential Averaging
§ Extreme cases:
§ Can only estimate the length
•
§ Based on length of previous CPU bursts,
using exponential averaging:
α =0
> τn+1 = τn
> Recent history does not count
•
1. t n = actual lenght of n th CPU burst
α =1
> τn+1 = α tn
2. τ n +1 = predicted value for the next CPU burst
> Only the latest CPU burst counts
3. α , 0 ≤ α ≤ 1
4. Define : τ n+=1 = α t n + (1 − α )τ n .
§ Otherwise: Expand the formula:
τn+1 = α tn + (1 - α)α tn-1 + …
+(1 - α )j α tn-j + …
+(1 - α )n +1 τ0
•
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
© Silberschatz, Galvin and Gagne 2005
6.13
Priority Scheduling
Since both α and (1 - α) are less than or equal to 1,
each successive term has less weight than its predecessor
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.14
© Silberschatz, Galvin and Gagne 2005
Round Robin (RR)
§ A priority value (integer) is associated with each process
§ Each process gets a small unit of CPU time:
§ The CPU is allocated to the process with the highest priority
(smallest integer ≡ highest priority)
• time quantum, usually 10-100 milliseconds.
§ After this time has elapsed, the process is preempted
and added to the end of the ready queue.
§ Given n processes in the ready queue and time quantum q,
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.
§ Performance
•
•
preemptive
nonpreemptive
§ 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
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
© Silberschatz, Galvin and Gagne 2005
6.15
Example: RR with Time Quantum q = 20
Process
P1
P2
P3
P4
Burst Time
53
17
68
24
•
•
•
q very large ⇒ FCFS
q very small ⇒ many context switches
q must be large w.r.t. context switch time,
otherwise too high overhead
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.16
© Silberschatz, Galvin and Gagne 2005
Time Quantum and Context Switches
Smaller time quantum ⇒ more context switches
§ The Gantt chart is:
P1
0
P2
20
P3
37
P4
57
P1
77
P3
P4
P1
P3
P3
97 117 121 134 154 162
§ Typically, higher average turnaround than SJF, but better response
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.17
© Silberschatz, Galvin and Gagne 2005
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.18
© Silberschatz, Galvin and Gagne 2005
3
RR: Turnaround Time Varies With Time Quantum
RR: Turnaround Time Varies With Time Quantum
Time Quantum = 1
Time Quantum = 2
P1
P2
P3
P4
P1
P2
P3
P4
3 + 9 + 15 + 17
=11
4
3
5
9 10
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
15
55
17
6.19
© Silberschatz, Galvin and Gagne 2005
RR: Turnaround Time Varies With Time Quantum
10
10
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
14 15
17
The average
turnaround time can
5 +10 + 14
17
be improved
if +most
=11.5
4
processes finish
their
next CPU burst in a
single time quantum.
quantum.
6.20
© Silberschatz, Galvin and Gagne 2005
Problems with RR and Priority Schedulers
§ Priority based scheduling may cause starvation
for some processes.
§ Round robin based schedulers are maybe too ”fair”...
we sometimes want to prioritize some processes.
The average
turnaround time can
in general
be improved if most
processes finish
their next CPU burst
in a single time
quantum.
quantum.
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.21
© Silberschatz, Galvin and Gagne 2005
Multilevel Queue
6.22
© Silberschatz, Galvin and Gagne 2005
§ Fixed priority scheduling
§ Ready queue is partitioned into separate queues, e.g.:
•
Serve all from foreground queue, then from
background queue.
•
Possibility of starvation.
§ Time slice
foreground (interactive)
background (batch)
§ Each queue has its own scheduling algorithm
•
•
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
Scheduling between the queues
Useful when processes are easily classified into
different groups with different characteristica...
characteristica...
•
•
§ Solution: Multilevel queue scheduling ...?
foreground – RR
•
Each queue gets a certain share of CPU time
which it can schedule amongst its processes
•
Example: 80% to foreground in RR, 20% to
background in FCFS
background – FCFS
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.23
© Silberschatz, Galvin and Gagne 2005
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.24
© Silberschatz, Galvin and Gagne 2005
4
Multilevel Queue Scheduling
Multilevel Feedback Queue
§ A process can move between the various queues
•
aging can be implemented this way
§ Time-sharing among the queues in priority order
•
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.25
© Silberschatz, Galvin and Gagne 2005
•
•
high
Q1 – RR with q = 16 ms
Q2 – FCFS
§ Scheduling:
•
•
•
•
A new job enters queue Q0 which is served RR.
When it gains CPU, the job receives 8 milliseconds.
If it does not finish in 8 milliseconds, it is moved to Q1.
low
priority
At Q1 the job is again served RR
and receives 16 additional milliseconds.
• If it still does not complete, it is preempted and moved to Q2.
§ In this case high priority to processes with short CPU bursts
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.27
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.26
© Silberschatz, Galvin and Gagne 2005
§ Multilevel-feedback-queue scheduler
defined by the following parameters:
•
•
•
•
•
number of queues
•
priority level of each queue
scheduling algorithms for each queue
method used to determine when to upgrade a process
method used to determine when to demote a process
method used to determine which queue a process will
enter when that process needs service
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.28
Multiprocessor Scheduling
Multiprocessor work sharing
§ CPU scheduling more complex when multiple CPUs are
available
§ One centralized task queue
•
•
•
© Silberschatz, Galvin and Gagne 2005
Multilevel Feedback Queue
Example of Multilevel Feedback Queue
§ Three queues:
• Q0 – RR with q = 8 ms
Processes in lower queues get CPU only if higher
queues are empty
Multiprocessor (SMP): homogeneous processors
Multi-core processors, CMP
§ CPU local task queues
§ Processor affinity
•
•
Multithreaded processors
> HW-Contextswitch-on-Load
© Silberschatz, Galvin and Gagne 2005
Decrease cache penalty
Resource only available on one CPU
§ Load balancing (push/pull migration)
> Cycle-by-cycle interleaving
•
•
> Simultaneous multithreading
OS task monitor and push work to less busy CPU
Low load CPU pull work from busy CPU
§ Supercomputing applications often use a fixed-sized process
configuration (“SPMD”, 1 thread per processor) and
implement own load balancing methods
§ More in TDDC78 ...
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.29
© Silberschatz, Galvin and Gagne 2005
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.30
© Silberschatz, Galvin and Gagne 2005
5
Real-Time Scheduling
Scheduling Algorithm Evaluation 1
§ Hard real-time systems
• required to complete a critical task within a guaranteed
amount of time
• missing a deadline can have catastrophic consequences
§ Deterministic modeling
•
•
§ Soft real-time computing
•
•
requires that critical processes receive priority
over less important ones
missing a deadline leads to degradation of service
> e.g., lost frames / pixelized images in digital TV
§ Often, periodic tasks or reactive computations
• require special scheduling algorithms: RM, EDF, ...
Type of analytical evaluation
takes a particular predetermined workload
and defines the performance of each algorithm for that
workload.
§ Queuing models
•
Little’s formula:
queue = arrival_rate x waiting_time
•
•
Only approximations
Questionable accuracy
Separate lecture on Real-Time Operating Systems
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.31
© Silberschatz, Galvin and Gagne 2005
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
Scheduling Algorithm Evaluation 2
Summary
§ Simulation Models
• Software system model
§ CPU Scheduler and Dispatcher
•
•
•
Process generation:
> Mathematical distribution
> Trace tape from real system
•
•
•
•
CPU utilization, throughput, ...
Preemptive vs Non-preemptive scheduling
RR, FCFS, SJF
Priority scheduling
Multilevel queue and Multilevel feedback queue
§ Multiprocessor Scheduling
§ (Realtime Scheduling: see Lecture on realtime-OS)
§ In the book (Chapter 5):
Scheduling in Solaris, Windows XP, Linux
High cost
6.33
Enable multiprogramming
§ Scheduling Algorithms
§ Implementation
• What is more accurate
than implementing the
scheduler in a real OS and test it?
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
© Silberschatz, Galvin and Gagne 2005
§ Goals:
=> Run simulation
•
6.32
© Silberschatz, Galvin and Gagne 2005
TDDI04, K. Arvidsson, IDA, Linköpings universitet.
6.34
© Silberschatz, Galvin and Gagne 2005
6