Scheduling - Hong Kong Polytechnic Universitycsajaykr/myhome/file/Scheduling.pdf · Load Balancing Symmetric Multithreading Algorithm Evaluation Real Time Scheduling Scheduling Examples

EEL 358 1

Scheduling

Reading:Silberschatzchapter 6

Additional Reading:Stallingschapter 9

EEL 358 2

OutlineIntroductionTypes of SchedulingScheduling CriteriaFCFS SchedulingShortest-Job-First SchedulingPriority SchedulingRound Robin SchedulingMultilevel Queue SchedulingMultiprocessor Scheduling

Load BalancingSymmetric Multithreading

Algorithm EvaluationReal Time SchedulingScheduling Examples

Windows XP, 2000Linux

EEL 358 3

Basic PointsProcess Scheduling, Thread Scheduling

Max CPU Utilization → Multiprogramming

CPU Burst ↔ I/O Burst

CPU Burst Distribution

Introduction

EEL 358 4

Histogram – CPU Burst Duration

EEL 358 5

CPU Scheduling DecisionsRunning → Waiting state

e.g. I/O Request, wait by ParentRunning → Ready state

e.g. InterruptWaiting → Ready state

e.g. Completion of I/OProcess Termination

Nonpreemptive SchedulingPreemptive Scheduling

Associated CostDesign of OS Kernel

Process → Kernel, wait for sys call or I/O completion beforecontext switch

Scheduling Types

EEL 358 6

Scheduling CriteriaCPU Utilization – How busy is the CPU?

Throughput – Number of processes that are completed per unit time

Turnaround Time – How long to execute a process? Submission ↔ Completion

Waiting Time – Sum of periods spent in ready queue

Response Time – Process Request → First response

EEL 358 7

Optimization CriteriaMax CPU utilization

Max throughput

Min turnaround time

Min waiting time

Min response time

Conflicting goals! Requires careful balanceAverage, Min/Max, Variance

EEL 358 8

Process Burst TimeP1 24P2 3P3 3

Arrivals in the order: P1 , P2 , P3 The Gantt Chart for the schedule:

Waiting time → P1 = 0; P2 = 24; P3 = 27Average waiting time → (0 + 24 + 27)/3 = 17

P1 P2 P3

24 27 300

FCFS Scheduling

EEL 358 9

FCFS SchedulingSay, if the processes arrive in the order

P2 , P3 , P1

The Gantt chart for the schedule:

Waiting time → P1 = 6; P2 = 0; P3 = 3Average waiting time → (6 + 0 + 3)/3 = 3, Previously 17 ↑Convoy effect → Short process behind long processNonpreemptive → Problem for time sharing systems

P1P3P2

63 300

EEL 358 10

SJF SchedulingCPU assigned to process with smallest next CPU burst, Tie → FCFS

Shortest-next-CPU-burst algorithm

Major difficultyEstimating the processing time of each job, Predicting the Next!Long running jobs may starve, steady supply of short jobs to CPU

SJF is optimal – minimum average waiting time

EEL 358 11

Process Arrival Time Burst TimeP1 0.0 7P2 2.0 4P3 4.0 1P4 5.0 4

SJF (non-preemptive)

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

SJF Scheduling

P1 P3 P2

73 160

P4

8 12

EEL 358 12

SJF SchedulingProcess Arrival Time Burst Time

P1 0.0 7P2 2.0 4P3 4.0 1P4 5.0 4

SJF (preemptive)

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

P1 P3P2

42 110

P4

5 7

P2 P1

16

EEL 358 13

Priority SchedulingA priority number (integer) associated with each process

SJF – A Priority schedulingEqual Priority - FCFS

CPU → Process with the highest priority, High ↔ LowPreemptiveNonpreemptive

Defining PrioritiesInternally, Measurable Quantities

Memory required, time limits, # open files, ratio of avg I/O to CPU burst, etc.

Externally, Outside OSImportance of Process, type/amount of funds, etc.

StarvationLow priority processes may never execute

Solution?Aging

EEL 358 14

Round Robin (RR)Each process gets a small unit of CPU time

Time Quantum (time-slice)usually 10-100 milliseconds

Time elapsed → PreemptedIf not completed → end of the ready queue

RR reduces penalty for short jobs in FCFS

Critical Issue → Length of quantum, qq large → FIFO or FCFSq small → Context switch overhead

EEL 358 15

Process Burst TimeP1 53P2 17P3 68P4 24

The Gantt chart is:

Typically, higher average turnaround than SJF, but better response.

P1 P2 P3 P4 P1 P3 P4 P1 P3 P3

0 20 37 57 77 97 117 121 134 154 162

Round Robin (RR)

EEL 358 16

Round Robin (RR)

EEL 358 17

Round Robin (RR)

EEL 358 18

Multilevel QueueReady queue → separate queues

foreground (interactive)background (batch)

Each queue → own scheduling algorithm, e.g.foreground – RRbackground – FCFS

Scheduling must be done between the queuesFixed priority scheduling; (i.e., serve all from foreground then from background), StarvationTime slice – each queue gets a certain amount of CPU time which it can schedule amongst its processes; i.e., 80% to foreground in RR20% to background in FCFS

EEL 358 19

Multilevel Queue Scheduling

EEL 358 20

Multilevel Feedback QueueSeparate processes → CPU burst characteristicsProcess moves up ↔ down in queues

Too much time ↓Aging ↑

Key pointsnumber of queuesscheduling algorithms for each queuemethod used to determine when to upgrade a processmethod used to determine when to demote a processmethod used to determine which queue a process will enter when that process needs service

EEL 358 21

Example of Multilevel Feedback Queue

CPUPrimary CPUScheduling

EEL 358 22

Multiple-Processor SchedulingMultiple CPUs → High scheduling complexityHomogeneous Processors

Asymmetric MultiprocessingNo data sharing, System data structures → one processor

Symmetric MultiprocessingSelf Scheduling, Ready queue

Processor AffinitySoft Vs Hard affinity

Load BalancingPush MigrationPull Migration

EEL 358 23

Algorithm EvaluationDeterministic modeling

Takes a particular predetermined workload and defines the performance of each algorithm for that workload

Queueing modelsQueue of network serversLittle’s formula, l = λ × w

λ - Avg arrival rate, w - Avg waiting time, l - Avg queue length

SimulationModel, clockSimulation → modifies system with clock ↑Distribution driven simulationOnly # instances of an event, order?

EEL 358 24

Evaluation of CPU Schedulers by Simulation

EEL 358 25

Real-Time SchedulingHard real-time systems

Complete a critical task within a guaranteed timeAdmit or RejectImpossible with SS, VMResource Reservation

Soft real-time computingCritical processes receive priority over less fortunate onesGeneral-purpose systems → Multimedia, GraphicsPriority InversionPriority-inheritance protocol

EEL 358 26

SchedulingPriority-based, preemptive schedulingThread runs → preempted by higher priority thread, terminates, QuDoes not guarantee execution of a real-time thread within time-limit

Thread Priorities32 level priority schemeReal time class → 16-32Variable class → 1-15Memory Management → Thread at 0 priority

Six Classes (Win32 API) – 1 + 5Within each 6 classes – 7 relative prioritiesCurrently selected foreground process → Scheduling Quantum ↑ 3

Example: Windows XP, 2000

EEL 358 27

Windows XP, 2000 Priorities

Relative Priority ↓

Priority Classes →

← Base Priority

EEL 358 28

Example: LinuxScheduling

Increased support for SMP, Scaling with # tasksProcessor affinity, load balancingHigh priority tasks → longer quanta, vice-versaReal time tasks – static prioritiesRest dynamic → nice values ± 5 (interactivity)

Numeric Priorities0-140 level priority schemeReal time → 0-99Nice values → 100-140