operating systems iii. scheduling - eurecomsoc.eurecom.fr/os/docs/courseos_iii_scheduling.pdfi...

Operating Systems

III. Scheduling

Ludovic [email protected], office 470

http://soc.eurecom.fr/OS/

@OS Eurecom

http://soc.eurecom.fr/OS/

Basics of SchedulingScheduling algorithms

Scheduling in Windows, Linux and Android

DefinitionsSwitching Between Processes

Outline

Basics of SchedulingDefinitionsSwitching Between Processes

Scheduling algorithms


2/35 Fall 2017 Institut Mines-Telecom Operating Systems - Scheduling




Three-level Scheduling: Admission, CPU andMemory





Three-Level Scheduling: Explanations

I Admission schedulerI Jobs are first stored into an admission queueI Jobs are then admitted in the system

I Memory schedulerI Swap in / swap outI To be avoided: Disk storage is expensive in terms of timeI For example, good to keep in memory 5 processes which

compute 20% of their time, and store all the others processeson disk

I CPU schedulerI See next slides

I In interactive systems, usually only 2 levels of schedulerI Memory and CPU





Scheduling Criteria

All systems

I Fairness: Giving each process a fair share of the CPU

I Policy enforcement: No process should be able to overcomethe rules of the system

I Balance: Keeping all parts of the system busy

Batch/Cloud systems

I CPU utilization: Keep the CPU busy

I Throughput: Maximize the number of processes that arecompleted per time unit (hour)

I Turnaround Time: Minimize time between submission andtermination





Scheduling Criteria (Cont.)

Interactive systems

I Response time: Respond to interactions as fast as possible

I Proportionality: Meet users’ expectations

Real-time systems

I Meeting deadlines: Keep the CPU busy

I Predictability: Know in advance whether time constraintscan always be met, or not





When to Schedule?

I At creation of a processI Run parent or child?

I At termination of a process

I When a process blocks on I/O, semaphore, etc.

I When I/O interrupt occursI At hardware clock interrupt

I Non-preemptive schedulingI The same process is given the CPU until it blocks or

voluntarily releases the CPUI Only choice if no timer is availableI Windows 3.1

I Preemptive schedulingI The same process can run only for a pre-defined time-intervalI The OS sets a timer to the time-interval





Non-Preemptive Scheduler





Preemptive Scheduler





Scheduling Implementation




Batch systemsInteractive systemsMultiprocessor systems

Outline

Basics of Scheduling

Scheduling algorithmsBatch systemsInteractive systemsMultiprocessor systems






Presented Scheduling Algorithms

For batch systems

I First-Come First-Served

I Shortest Job First

I Shortest Remaining TimeNext

For interactive systems

I Round-Robin scheduling

I Priority-based scheduling

I Group-based scheduling

I Guaranteed scheduling

I Fair-share scheduling

I Lottery scheduling

For multiprocessors systems

Presentation of three differentapproaches

For RT systems

Attend lectures on RTOS to knowmore!





First-Come, First-Served (FCFS)

Processes are assigned the CPU in the order they request it

I Single queue of ready processes

I Easy to program, fair

I Non-preemptive scheduling

Example: Average Wait Time for various sets of processes

Process p1 p2 p3

Duration 24 3 3

Arrivaltime (@)

0 0 0

Arrival order AWT=?

p1, p2, p3 0+24+273 = 17

p2, p3, p1

p3, p2, p1





Shortest Job First (SJF)

Scheduler selects the job with the shortest computation time

I Easy to implement

I Non-preemptiveI Optimal when processes are all ready simultaneously

I Minimum AWT

But ... How to predict the next CPU burst time???

1. Specified amount

2. Exponential average

τn+1 = αtn + (1− α)τn

I τ : predicted valueI t: measured computation time





Shortest Job First: Example

Example ]1: Same arrival time

Process p1 p2 p3 p4

Duration 8 4 4 4

@ 0 0 0 0

Algo AWT=?

FCFS (p1,p2, p3, p4)

SJF

Example ]2: Various arrival times

Process p1 p2 p3 p4 p5

Duration 2 4 1 1 1

@ 0 0 3 3 3

Algo AWT=?

FCFS

SJF

Can you find a better scheduling???





Shortest Remaining Time Next (SRTN)

Preemptive version of Shortest Job First

Memo: SJF is non-preemptive

Example: Comparison between SJF and SRTN

Process p1 p2

Duration 10 1

@ 0 1

Algo Scheduling AWT=?

SJF p1 for 10, p2 for 1 0+92 = 4.5

SRTN p1 for 1, p2 for 1, p1 for 9 1+02 = 0.5





Round-Robin (RR)

Each process is assigned a time quantum

I If the process is still running at the end of its quantum, it ispreempted

I Next process is assigned the CPU

I Widely used, easy to implement, fair

How to set the value of the quantum???

I Quantum too short? Quantum too long?I What happens if:

I quantum = ε?I quantum =∞?

I Typical quantum: 10 to 50 ms





Round-Robin: Explanations





Round-Robin: Example

Process p1 p2 p3

Duration 24 3 3

@ 0 0 0

Quantum 4 4 4

p1 p2 p3 p1 p1 p1 p1 p1

0 4 7 10 14 18 22 26

AWT=???





Priority-Based Scheduling

Limitations of RRI RR assumes all processes are of equal importance

I For example, sending of an email vs. playing music

Priority-based scheduling: Priorities are assigned to processes

I Static priority or dynamic priorityI The process with the higher priority (may be the lower value)

is chosenI RR between processes of the same priority





Priority-Based Scheduling: Dynamic Priorities

Priorities might be re-evaluated

I Avoids high-priority processes to jeopardize the CPU(starvation of low-priority processes)

I For example, priority can be set to priorityn+1 = quantumtn

where tn is the last computation time

Example: 10 ms quantum

I Process used 1ms → new priority = 10

I Process used 5ms → new priority = 2

I Process used 10ms (i.e., all quantum) → new priority = 1





Priority-Based Scheduling (Cont.)





Group-based

Scheduling policy is applied accordingto groups of processes

Example: group 1 may have a quantumof 20, and group 2 a quantum of 10

Intra-group scheduling

All algorithms previously presented can be used

Inter-group scheduling

I Fixed-priority preemptive schedulingI Highest priorities for foreground processes (interactivity)

I Time-slice between groupsI e.g., 80% for group 1, 20% for group 2





Guaranteed Scheduling

I Special case of ”Group-based”

Idea: Give a ratio of the CPU to groups of processes

I For example, users are given a fraction of CPU (e.g., 1n )

I Compute the ratio between actual CPU time consumed andentitled CPU time

I Run the process with the lowest ratio until its ratio has movedabove its closest competitor





Fair-Share Scheduling

Idea: Fairness between users is taken into account firstI For example, in RR, a user with 9 processes gets more CPU

than a user with 2I Allocates a fraction of CPU to users

I One user = one scheduling group

I Other possible notions of fairnessI Resources, etc.





Lottery Scheduling

Idea: Lotto tickets are given to processes

I A random ticket is picked up, the winner gets a quantum oftime

I The number of tickets received by a process is equivalent toits importance

I Processes can exchange tickets for cooperation

I Highly responsive

Example

A video server with three video streams at 10, 20, 25 frames / sec,respectively.→ Each process is given a number of tickets equals to the framerate i.e 10, 20 and 25


44

9

3

8

33

3




Multiprocessor Scheduling

One scheduling queue for each processor

I Symmetric system

A common ready-queue for all processors

I Asymmetric multi-processing

I Danger: synchronization problems because several processorscould access to the scheduling data structures at the sametime

Only one ready-queue accessed by only one processor

I No synchronization issues

I Drawback?




WindowsSolaris, LinuxAndroid

Outline

Basics of Scheduling

Scheduling algorithms

Scheduling in Windows, Linux and AndroidWindowsSolaris, LinuxAndroid





Windows NT Scheduling

BasicsI Priority-based preemptive round-robin scheduling

I Raises the priority of interactive and I/O bound processes

I CPU cycle-based scheduling (since Vista)

A process runs until ...

I It is preempted by a higher priority process

I It terminates

I Its time quantum expires (12 ms in Windows 2000 - highervalue is used in ”server” versions)

I It calls a blocking system call





Windows NT: Priority Classes





Windows NT: Changing Classes of Processes





Solaris 10: Kernel Architecture


source = Oracle/Sun Microsystems




→ Linux 2.6.22: Priority-Based Scheduler

Scheduling classes

I Real-time processes: SCHED FIFO, SCHED RR

I Interactive and batch processes: SCHED OTHER,SCHED BATCH

I Low-priority processes: SCHED IDLEI One active queue for each of the 140 priorities and for each

processorI Cross-CPU scheduling regularly performed (e.g., every 200 ms)

SCHED OTHER

Round-Robin time-sharing policy with dynamic priorities

I Processes running for a long time are penalized





Linux 2.6.23 → . . . : Completely FairScheduler

I ”Out of balance” tasks are given time to executeI Out of balance task = task has not been given a fair amount

of time relative to others

I Preemption time is variable: it depends upon the time balanceof the task w.r.t. other tasks

I The amount of time provided to a given task is called thevirtual runtime

I Group-based: the virtual runtime can also be computed for awhole group

I Priorities are used as a decay factor for the time a task ispermitted to execute





Android

I Roughly, the scheduler is based onthe Linux one

I → Fair scheduling approach

I BUT: fairness according to Groupsof processes

I Foreground/Active, visible,service, background, empty

I To reclaim resources, Android maykill processes according to theirrunning priority

Foreground

Visible

Service

Background

Empty

High priority

Low priority


operating systems iii. scheduling - eurecomsoc.eurecom.fr/os/docs/courseos_iii_scheduling.pdfi...

Documents