Operating Systems ICS 431

Weeks 6 - 7
Ch. 5: Processes Scheduling
Dr. Tarek Helmy El-Basuny
Dr. Tarek Helmy, ICS-KFUPM
 Ch: 5 Process Scheduling
• Schedulers Overview (Short, Medium and Long)
• Time Quantum/Burst Time and Context Switch
• Scheduler, Dispatcher and Swapper modules in the OS
• Basic Assumptions for Process Scheduling
• Types of Short-term Scheduling Algorithms (Preemptive, and Non-preemptive)
– Scheduling decision and the Types of processes
• Evaluation Criteria of Scheduling Algorithms
• Scheduling Algorithms
– First-Come, First-Served (FCFS)
– Shortest Job First (SJF)
– Shortest Remaining Time First (SRTF)
– Process Burst Length Prediction
– Round Robin
– Priority-Based
– Multilevel Queue
– Multilevel Feedback Queue
• Scheduling Algorithms Evaluation
– Deterministic and Queuing models
– Simulations, and Implementation
• Scheduling Polices in Different OS
• Multiprocessor Systems: Just Introduction, Scheduling of Multiprocessors systems

Dr. Tarek Helmy, ICS-KFUPM 2

 Ch: 5 (Process Scheduling) Objectives

• Recognize the importance of CPU scheduling for the operation

of computer systems.

• Understand the decision policy of different CPU scheduling

algorithms.

• Know about the strengths and weaknesses of each algorithm.

• Obtain the knowledge to evaluate and select the appropriate

scheduling algorithm for different computing environments.

Dr. Tarek Helmy, ICS-KFUPM 3

 Schedulers Overview
• Processes entering the system are put into a job queue.
• Processes ready to be executed are kept in the ready queue.
• Processes waiting for a particular device are placed in one of the I/O queues.
Medium-Term Scheduler
Ready
New Processes (Job Running
Queue Suspended
Queue)

Long-term Short-term Scheduler

Scheduler
• Scheduler: a module decides which process will be dispatched to the CPU.
• Dispatcher: a module gives control of the CPU to the process selected by the short-term
scheduler.
• Swapper - manages transfer of processes between main memory and virtual memory
(medium-term scheduler)
• Dispatch/Context latency: Time it takes for the OS to stop one running process and to start
running another one.
• In the Context/Dispatch time, the OS will:
• Remove the “old” process from the CPU by the dispatcher.
• Select the next process by the scheduler.
• Allocate the “new” process to the CPU by the dispatcher.
Dr. Tarek Helmy, ICS-KFUPM 4
 Time Quantum/Burst Time

• How does the OS interleave execution of many processes on the CPU?

• Answer: By sharing the CPU time among the processes
• Time Quantum/Burst Time: Amount of CPU time given to each process.
Q: What happens if the quantum time is too short?
A: Too much overhead for context switching.
Q: What if the time quantum time is too long?
A: Poor response times, some processes will starve
A Context Switch occurs when a process exchange is made between the ready
and run states.

Ready Run

Dr. Tarek Helmy, ICS-KFUPM 5

 During the Context Switch from Process A to B

1. OS Kernel:
• Sets the PC register with the address of the Interrupt Handler (IH),
which is an address in the interrupt vector of OS kernel.
• Setting the mode bit to switch from user mode to kernel mode.
• Copies A’s state from CPU registers to A’s PCB.
• Sets A’s state to Ready or Blocked.
• Inserts a pointer of A’s PCB on Ready-Queue or I/O-Queue.
2. Scheduler: Selects a process B to run, based on its scheduling policy.
3. OS Kernel:
• Sets B’s state to Running,
• Copies B’s state from B’s PCB to CPU registers.
• Kernel transfers control to B and switches from kernel mode back to
user mode.

Dr. Tarek Helmy, ICS-KFUPM 6

 Assumptions for Process Scheduling
Basic assumptions behind most scheduling algorithms:
• In a single processor system, only one process can be in running state at a time
• In a multi-core/multi-processor system, many processes can be in running state at a
time.
• In this course we are going to take care scheduling into a single processor.
• There is a pool of ready processes which compete for the CPU/CPUs or resources.
• The processes are either dependent/independent.
• In this course we are going to take care scheduling of independent processes.
• The job of the scheduler is to distribute the CPU time among different ready processes
''fairly'' (according to some definition of fairness) and in a way that optimizes some
performance criteria.
– i.e. to maximum CPU utilization , concurrent execution is recommended.
• Key to the success of multiprocessing scheduling:
– Process execution consists of cycles of CPU execution and cycles of I/O waiting.
• CPU-I/O bound processes:
– An I/O-bound process will have many I/O-bound instructions.
– A CPU-bound process will have many long CPU-bound instructions.

Dr. Tarek Helmy, ICS-KFUPM 7

 Sequence of CPU and I/O Bursts
• A process will run for a while (CPU burst), perform some IO (I/O burst), then
run for a while again (next CPU burst) and so on.
• The frequency of I/O and CPU operations depend on the process type.

load A
add A } CPU CPU Burst

read from file Wait for I/O I/O Burst

Increment A

write to file
} CPU
Wait for I/O
CPU Burst
I/O Burst

load A
add A
read from file
} CPU CPU Burst

Wait for I/O I/O Burst

Dr. Tarek Helmy, ICS-KFUPM 8

 Short-Term (CPU) Scheduler

• CPU Scheduler: is a module/process of the operating system

– Is responsible for the selection of the next process to be executed.

• The selection is done according to a scheduling algorithm

– The scheduler decision policy should be simple
– The scheduler uses resources of the computer system,
• In particular CPU time, memory
– The scheduler should not consume too much CPU time
• Otherwise the overhead is too high

Dr. Tarek Helmy, ICS-KFUPM 9

 Preemptive vs. Non-Preemptive Scheduler
Scheduling is Preemptive:
• If the CPU can be taken away from a process during its execution due to:
• Occurrence of interrupts
• Arrival of higher priority process
• A change of status
• Time limit
• Prevents a process from using the CPU for too long
• May lead to race conditions which will be solved by using process synchronization
• Supported by Windows, UNIX, Linux, Mac-OS 8 for the PowerPC platform
Scheduling is non-Preemptive:
• If, once the CPU has been allocated to a process, the process can keep the CPU
until:
• It releases the CPU, either by terminating or switching to the waiting state.
• Supported by old versions of Microsoft Windows and Mac OSs.
• Simple, and easy to implement
• May lead to starvation
• Not suited for multi-user systems as it decreases the responsiveness.

Dr. Tarek Helmy, ICS-KFUPM 10

 Scheduling Decision for Processes

• Batch Process
– Users are not waiting at their terminals for a response time.
– Should scheduling be preemptive or non-Preemptive?
• Since a batch process does not need any responsiveness, the OS
can use non-preemptive and save the cost of context switch.

• Interactive Process
– Environment has interactive users
– Should scheduling be preemptive or non-Preemptive?

• Real Time Process

– Processes may have real-time constraints
– Should scheduling be preemptive or non-Preemptive?
• Since a real-time process needs more responsiveness, so it is
recommended to be scheduled preemptively.

Dr. Tarek Helmy, ICS-KFUPM 11

 CPU or Short-Term Scheduler

• Selects one of the processes in the

ready queue to be allocated to the New Terminated

CPU.
• Scheduling decisions may take Admit Exit
Scheduler
place when a process: Dispatch

1. Switches from running to waiting

state. ex. I/O request. Ready
Running

2. Switches from running to ready

Timeout
state. ex. Due to interrupt, time
quantum expiring. Event
occurs
3. Switches from waiting to ready. Event
wait

ex. Completion of I/O.

4. Switches from new to ready. Waiting

5. Terminates.
– Scheduling under 1 and 5 are
non-preemptive while 2, 3 and 4
are preemptive.

Dr. Tarek Helmy, ICS-KFUPM 12

 Evaluation Criteria of Scheduling Algorithms

• Fairness
– Does the algorithm fairly share the CPU time among processes?
• CPU utilization: The fraction of time a CPU is executing processes. (ratio of in-use
time/total observation time). Note that CPU might be busy but in context switching.
• Resources utilization: Efficient utilization of all resources, are all recourses used?
• Throughput: # of processes that complete their execution per a time unit.
• Turnaround Time: Elapsed time from the submission/arrival of a process to its
completion.
• Waiting Time: Amount of time a process has been waiting in the ready queue,
not in the waiting I/O queues, may consist of several separate periods.
• Response Time (interactive users): Amount of time it takes from the submission
until the first response is given to the process, not output (for time-sharing
environment).
• Context Switches: # of context switches: Indication for the amount of overhead
• Complexity of the scheduling algorithm
– Indication of the time needed to take a decision and select the next process.
– It also indicates the amount of memory going to be used by the algorithm.

Dr. Tarek Helmy, ICS-KFUPM 13

 Average Waiting, Turn around and Response Time

• Waiting Time: how much time the process spends on the ready queue
waiting to execute (this does not include time the process spends doing or
waiting for I/O).
– TW : Amount of time a process has been waiting in the ready queue, it
can be accumulated times, i.e. in round robin scheduling policy.
• Waiting Time starts when the process enters the ready queue (not when it
enters the system).
• Average Waiting Time (AWT of n processes) = Tw / n
• Response Time (of one process):
– TR= TStart to use CPU – T Arrive
• Average Response Time (ART of n processes) = TR / n
• Turnaround Time (of one process)
– TT = TFinish the process – T Arrive
• Average Turnaround Time (ATT of n processes) = TT / n
Dr. Tarek Helmy, ICS-KFUPM 14
 Ideal Process Scheduling Algorithm

• Should Maximize
– CPU and resources utilization
• Keep the CPU and resources busy all the time.
– Throughput
• # processes completed within a time unit.
• Should Minimize
– Waiting Time:
• Minimizes it leads to minimizing the Turnaround time.
– Turnaround Time
• Minimizes it leads to maximizing the throughput.
– Response Time
• Minimizes it leads to maximizing the user’s satisfaction.

Dr. Tarek Helmy, ICS-KFUPM 15

 Factors in Scheduling

• The scheduler determines the order in which processes are going to be

executed.
• Factors affect the scheduler’s decision and implementation:
– Which policy used to select a process? (will affect the response time)
– How long may the process keep the CPU? (time burst/quantum)
– How are conflicting requests resolved? (i.e. two processes have the
same priority or same arrival time) while applying the priority or FCFS
policies.
– Is the process CPU- or I/O-bound?
– Is the process interactive or batch?
– Execution time used so far (historical time bursts)
– Execution time required to complete (future time bursts)
– Preemption frequency (how many time a process has been preempted?)
– Page fault frequency (how many time a process has been swapped out
and in?)
– Processes dependency
– etc..

Dr. Tarek Helmy, ICS-KFUPM 16

 First-Come, First-Serve (FCFS) Scheduling Policy

• The process that arrives first is allocated the CPU first.

• Simply implement the Ready queue as a FIFO queue.
Example Process Burst Time
P1 24
P2 3
P3 3
• Arrival order: P1 , P2 , P3
– The Gantt Chart for scheduling is:
P1 P2 P3

0 24 27 30
– Waiting time for P1 = 0; P2 = 24; P3 = 27
– Tw= TStart to use CPU – T Arrive
• Average waiting time: (0 + 24 + 27)/3 = 17
• Do you think, changing the order of processes arrival will
affect the average waiting time and the throughput?

Dr. Tarek Helmy, ICS-KFUPM 17

 FCFS Scheduling Policy (Cont.)

• If the arrival order is: P2 , P3 , P1 .

– The Gantt chart for scheduling is: P BT
P2 3
P2 P3 P1
P3 3
0 3 6 30 P1 24

– Waiting time for P1 = 6; P2 = 0; P3 = 3

– TW= TStart to use CPU – T Arrive
• Average waiting time: (6 + 0 + 3)/3 = 3
• Much better than previous case.
• At time slot 6, the throughput will be 2 but in the
previous order, the throughput was 0, means more
productive than the previous order.
Dr. Tarek Helmy, ICS-KFUPM 18
 FCFS Policy Example 2
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

• Since the arrival time of all processes is 0 and all have the same priority, the order
of processes will be used to resolve the confliction.

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 19

 FCFS Policy Example 2
Proces s # A rriv al Time B urs t Length Prio rity
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

• Since the arrival time of all processes is 0 and all have the same priority, the order
of processes will be used to resolve the confliction.

P1

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 20

 FCFS Policy Example 2
Proces s # A rriv al Time B urs t Length Prio rity
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

• Since the arrival time of all processes is 0 and all have the same priority, the order
of processes will be used to resolve the confliction.

P1 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 21

 FCFS Policy Example 2
Proces s # A rriv al Time B urs t Length Prio rity
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

• Since the arrival time of all processes is 0 and all have the same priority, the order
of processes will be used to resolve the confliction.

P1 P2 P3

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 22

 FCFS Policy Example 2
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

• Since the arrival time of all processes is 0 and all have the same priority, the order
of processes will be used to resolve the confliction.

P1 P2 P3 P4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 23

 FCFS Policy Example 2
Proces s # A rriv al Time B urs t Length Prio rity
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

• Since the arrival time of all processes is 0 and all have the same priority, the order
of processes will be used to resolve the confliction.

P1 P2 P3 P4 P5

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 24

 FCFS Policy Example 2
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

• Since the arrival time of all processes is 0 and all have the same priority, the order
of processes will be used to resolve the confliction.

P1 P2 P3 P4 P5

0 5 10 15 20 25 30

Proces s # W aiting Time Res pons e Turnaround #of Context

Time Time S w itches
P1 0 0 6 1
P2 6 6 21 1
P3 21 21 24 1
P4 24 24 28 1
P5 28 28 30 1
Av erage 79/5 = 15.8 79/5 = 15.8 21.8 1

Dr. Tarek Helmy, ICS-KFUPM 25

 FCFS Policy Properties

• Very simple, easy to implement

– Uses a FIFO queue
– Very quick selection of the next process
• Constant time, independent of the number of processes in the
ready queue.
• Non-preemptive
• Often long average waiting and response times.
• Not suitable for multi-user systems!, Why?
• Since serving the short jobs first will improve the average waiting time
which in fact will improve the throughput.
• Why do not we sort the jobs based on their burst lengths and then serve
the shortest job first.

Dr. Tarek Helmy, ICS-KFUPM 26

 Shortest-Job-First (SJF) Policy

• Shortest Job First (SJF) or Shortest Job Next (SJN) or Shortest Process
Next (SPN): Select the waiting process with the smallest execution/burst time
to execute first.
• How do we know the burst length of each process?
• Predict for each process the length of CPU time it needs. One of its limitations.
• Use these predicted lengths to schedule the process with the shortest time
first.
• Shortest-Job-First (SJF):
– It could be Non-Preemptive: once CPU is given to the process, it cannot
be preempted until the processes completes its burst time.
– Do scheduling when a process finishes;
– Assume arrival time 0 for all processes!, practically is not.
– Assume the same priorities for all processes!, practically is not.
– Different (expected) burst lengths, how about if two have the same length.
– It could be Preemptive: upon arrival of a shorter process than the
running one, the CPU can be preempted and given to another process.
• SJF is optimal: it gives minimum average waiting time for a given set of
processes. Let us verify that.
Dr. Tarek Helmy, ICS-KFUPM 27
 SJF Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

WT= TStart to use CPU – T Arrive TT =TFinish the process – T Arrive

P5

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 28

 SJF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

WT= TStart to use CPU – T Arrive TT =TFinish the process – T Arrive

P5 P3

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 29

 SJF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

WT= TStart to use CPU – T Arrive TT =TFinish the process – T Arrive

P5 P3 P4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 30

 SJF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

WT= TStart to use CPU – T Arrive TT =TFinish the process – T Arrive

P5 P3 P4 P1

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 31

 SJF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

WT= TStart to use CPU – T Arrive TT =TFinish the process – T Arrive

P5 P3 P4 P1 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 32

 SJF Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 0 15 1
P3 0 3 1
P4 0 4 1
P5 0 2 1

WT= TStart to use CPU – T Arrive TT =TFinish the process– T Arrive

P5 P3 P4 P1 P2

0 5 10 15 20 25 30

Proces s # W aiting Time Res pons e Turnaround #of Context

Time Time S w itches
P1 9 9 15 1
P2 15 15 30 1
P3 2 2 6 1
P4 5 5 9 1
P5 0 0 2 1
Av erage 31/5 = 6.2 31/5 = 6.2 62/5 = 12.4 1

Dr. Tarek Helmy, ICS-KFUPM 33

 SJF: Different Arrival Times

• Slight modification

– Different arrival times for the processes

– Same priorities

– Different (predicted) burst lengths

– The processes waiting in the ready queue are added to the

diagram (select the shortest among processes waiting in the
ready queue). Non-arrived processes should not be counted
when deciding the shortest one.

Dr. Tarek Helmy, ICS-KFUPM 34

 SJF Example 1: Different Arrival Times
Process Arrival Time Burst Time
P1 0.0 7
P2 2.0 4
P3 4.0 1
P4 5.0 4
• TW= TStart to use CPU – T Arrive
AWT =( (0-0) + (8-2) + (7-4) + (12-5))/4 = 4
• TT = TFinish the job – T Arrive
ATT =((7-0) + (12-2) + (8-4) + (16-5))/4= 8
P1 P3 P2 P4

0 3 7 8 12 16

P2 P3 P4
arrives arrives arrives
Dr. Tarek Helmy, ICS-KFUPM 35
 SJF Example 2: Different Arrival Times
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 5 3 1
P4 8 4 1
P5 14 2 1

P1: 6 ready queue at time 0

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 36

 SJF Example 2: Different Arrival Times
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 5 3 1
P4 8 4 1
P5 14 2 1

P1: 6 P2: 15 ready queue at time 3

P1

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 37

 SJF Example 2: Different Arrival Times
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 5 3 1
P4 8 4 1
P5 14 2 1

P2: 15
P1: 6 P2: 15 P3: 3
ready queue at time 5

P1 P3

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 38

 SJF Example 2: Different Arrival Times
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 5 3 1
P4 8 4 1
P5 14 2 1

P2: 15 P2: 15
P1: 6 P2: 15 P3: 3 P4: 4

P1 P3 P4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 39

 SJF Example 2: Different Arrival Times
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 5 3 1
P4 8 4 1
P5 14 2 1

P2: 15 P2: 15
P1: 6 P2: 15 P3: 3 P4: 4 P2: 15

P1 P3 P4 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 40

 SJF Example 2: Different Arrival Times
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 5 3 1
P4 8 4 1
P5 14 2 1

P2: 15 P2: 15
P1: 6 P2: 15 P3: 3 P4: 4 P2: 15 P5: 2

P1 P3 P4 P2 P5

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 41

 SJF Example 2: Different Arrival Times
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 5 3 1
P4 8 4 1
P5 14 2 1

P2: 15 P2: 15
P1: 6 P2: 15 P3: 3 P4: 4 P2: 15 P5: 2

P1 P3 P4 P2 P5

0 5 10 15 20 25 30
Pro ces s # W aiting Time R es p o ns e Turnaro und # o f C o ntex t
Time Time S w itches
P1 0 0 6 1
P2 13-3 = 10 13-3 = 10 28-3 = 25 1
P3 6-5 = 1 6-5 = 1 9-5 = 4 1
P4 9-8 = 1 9-8 = 1 13-8 = 5 1
P5 28-14 = 14 28-14 = 14 30-14 = 16 1
A v erag e 26/5 = 5. 2 26/5 = 5. 2 50/5 = 10 1

Dr. Tarek Helmy, ICS-KFUPM 42

 SJF Optimality

long b short

short b long

• This graph Proofs that the SJF algorithms is optimal

• Better responsiveness
• Minimum average waiting time
• Improves the throughput
• What is the difficulties of theses algorithms?

Dr. Tarek Helmy, ICS-KFUPM 43

 SJF Policy Properties

• Much better average waiting time for processes when we use SJF
than FCFS.
– Optimal with respect to the average waiting time.
• Non-preemptive
• Relies on knowing the length of the CPU bursts
– In general, it is difficult to impossible to get it accurately.
• Implementation is more complex than FCFS
• Time of selecting a process is variant.
– Linear w.r.t. number of processes in the ready queue
• Impractical due to burst length prediction problem
• Starvation is possible
– If new, short processes keep on arriving, long processes may
never be served.
• How do we overcome this problem?

Dr. Tarek Helmy, ICS-KFUPM 44

 Ch: 5 Process Scheduling
• Schedulers Overview (Short, Medium and Long)
• Time Quantum/Burst Time and Context Switch
• Scheduler, Dispatcher and Swapper modules in the OS
• Basic Assumptions for Process Scheduling
• Types of Short-term Scheduling Algorithms (Preemptive, and Non-preemptive)
– Scheduling decision and the Types of processes
• Evaluation Criteria of Scheduling Algorithms
• Scheduling Algorithms
– First-Come, First-Served (FCFS)
– Shortest Job First (SJF)
– Shortest Remaining Time First (SRTF)
– Process Burst Length Prediction
– Round Robin
– Priority-Based
– Multilevel Queue
– Multilevel Feedback Queue
• Scheduling Algorithms Evaluation
– Deterministic and Queuing models
– Simulations, and Implementation
• Scheduling Polices in Different OS
• Multiprocessor Systems: Just Introduction, Scheduling of Multiprocessors systems

Dr. Tarek Helmy, ICS-KFUPM 4

 Shortest-Remaining-Time-First (SRTF) Policy

Shortest-Remaining-Time-First (SRTF)
– Preemptive: If a new process arrives with burst length less than
remaining time of current executing process, preempt and make a
context switch.
– Remaining time: (The process burst length - the time the CPU has
already spent serving this process).
• A scheduling decision must be made when
– A process is done with its CPU burst
– A new process arrives in the ready queue
• Arrival time of new processes is important
• It is important to keep track of the processes currently in the ready queue
– The policy used here: preempted processes go to the end of the ready
queue.
– It may depend on another policy for implementation.
Process Arrival Time Burst Time
P1 0.0 7
P2 2.0 4
P3 4.0 1
P4 5.0 4

Dr. Tarek Helmy, ICS-KFUPM 46

 Shortest-Remaining-Time-First

P1 P2 P3 P2 P4 P1

0 2 4 5 7 11 16

P2 arrives P3 arrives P4 arrives P2 finishes

RT: RT: (P3 finishes) RT:
P1:7-2= 5 P1: 5 RT: P1: 5
P2: 4 P2: 4-2=2 P1: 5 P4: 4
P3: 1 P2: 2
P4: 4

• AWT =  (TStart to use CPU – T Arrive )/n=(9 + 1 + 0 +2)/4 = 3

• ATT =  (TFinish the job – T Arrive)/n=(16+5+1+6)/4 = 7

Dr. Tarek Helmy, ICS-KFUPM 47

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 1

P1: 6

P1

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 48

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 1

P1: 3
P1: 6 P2: 15

P1 P1

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 49

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 1

P1: 3
P1: 6 P2: 15 P2: 15

P1 P1 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 50

 SRTF Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 1

P1: 3 P2: 12
P1: 6 P2: 15 P2: 15 P3: 3

P1 P1 P2 P3

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 51

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 1

P1: 3 P2: 12
P1: 6 P2: 15 P2: 15 P3: 3 P2: 12

P1 P1 P2 P3 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 52

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 1

P1: 3 P2: 12 P2: 10

P1: 6 P2: 15 P2: 15 P3: 3 P2: 12 P4: 4

P1 P1 P2 P3 P2 P4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 53

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 P4: 1 1

P1: 3 P2: 12 P2: 10 P5: 2

P1: 6 P2: 15 P2: 15 P3: 3 P2: 12 P4: 4 P2: 10

P1 P1 P2 P3 P2 P4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 54

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 P4: 1 1

P1: 3 P2: 12 P2: 10 P5: 2 P5: 2

P1: 6 P2: 15 P2: 15 P3: 3 P2: 12 P4: 4 P2: 10 P2: 10

P1 P1 P2 P3 P2 P4 P5

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 55

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 P4: 1 1

P1: 3 P2: 12 P2: 10 P5: 2 P5: 2

P1: 6 P2: 15 P2: 15 P3: 3 P2: 12 P4: 4 P2: 10 P2: 10 P2: 10

P1 P1 P2 P3 P2 P4 P5 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 56

 SRTF Policy Example

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 P4: 1 1

P1: 3 P2: 12 P2: 10 P5: 2 P5: 2

P1: 6 P2: 15 P2: 15 P3: 3 P2: 12 P4: 4 P2: 10 P2: 10 P2: 10

P1 P1 P2 P3 P2 P4 P5 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 57

 SRTF Policy Properties

• Good response time for short processes

– Attractive for multi-user systems
• Preemptive version of the SJF algorithm
– Higher overhead (context switches)
• Starvation possible
• Impractical due to burst length prediction problem

Dr. Tarek Helmy, ICS-KFUPM 58

 FCFS, SJF, SRTF Policies Comparison

• With the same set of processes, we can compare the performance of the
FCFS, SJF and SRTF.
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 9 3 1
P4 14 4 1
P5 17 2 1

FCFS SJF SJF (different arrival times) SRTF

AWT 15.8 6.2 5.2 2.6

ART 15.8 6.2 5.2 2.6

ATT 21.8 12.4 10 7.2

• What does the lowest AWT, ART, and ATT mean?

• Is it supporting the optimality of the algorithms?
• What is the main headache of this algorithm?

Dr. Tarek Helmy, ICS-KFUPM 59

 Burst Length Prediction

• Practically, the CPU burst length of a process is not known and needs
to be predicted/estimated.
– Algorithms such as SJF, SRTF in their pure form can’t be used in
practical systems unless there is an algorithm to predict the next
process burst length. How?
• The past can be a good predictor of the future, moreover
– Recent bursts might be given more importance than older bursts.
• The CPU burst length can be estimated based on the previous CPU
bursts of the process.
• This requires analysis of the code to be executed while the
scheduling decision is made.
• That means, additional overhead during the scheduling decision
– Time to estimate next CPU burst of the process and calculate the remaining time.
– Memory space to keep values of recent CPU burst lengths.

Dr. Tarek Helmy, ICS-KFUPM 60

 Burst Length Prediction

• A generic formula to estimate the length of the next CPU burst from the
lengths of previous ones is:
• E i = * A i-1 + (1- ) * E i-1
• Ei: estimated burst length of a process at time i
• Ai - 1: actual burst length of a process at time (i -1)
• Ei-1: previously estimated burst of a process at time (i -1)
• : factor to balance the importance of recent and not so recent bursts
– If  =0 then E i = E i-1
– If  =1 then E i = A i -1
• At time ti we estimate the length of the next CPU burst based on information
we have about previous bursts according to:
E i =  * A i -1 + (1- ) * E i -1
– We have to select an initial value for  (here 0.75)
– For the first burst A0, we randomly set it, after that we use the measured
time for the previous burst A i - 1.
– The measured burst time may be significantly different from the estimate.

Dr. Tarek Helmy, ICS-KFUPM 61

 Burst Length Prediction: Example

E i =  * A i -1 + (1- ) * E i -1

Time Estimate Actual Burst

(say 10)
T0 0.75 * 10 + 0.25 * 0 = 7.5 6
T1 0.75 * 6 + 0.25 * 7.5 = 6.375 3
T2 0.75 * 3 + 0.25 * 6.375 = 3.85 7
T3 0.75 * 7 + 0.25 * 3.85 = 6.2 12
T4 0.75 * 12 + 0.25 * 6.2 = 9.55

Dr. Tarek Helmy, ICS-KFUPM 62

 Example of Predicting CPU Bursts

Actual
12
10
burst length

Predicted
8
6
4
2

time

Actual CPU Burst Ai-1 6 4 6 4 12 12 .. 12

W=1/2 .
Predicted Ei 10 8 6 6 5 8.5 10 .. 11
.
Dr. Tarek Helmy, ICS-KFUPM 63
 Round Robin (RR) Policy

• It is a preemptive scheduling algorithm.

• Each process takes an equal or un-equal share
of CPU time (time quantum) in turn. After this
time has elapsed, the process is preempted and
added to the end of the ready queue.
• 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.
• Performance
– q time large  FCFS (most of the processes
will finish in that quantum time)
– q time small  more context switches,
overhead is too high and throughput is low.
– q time must be large with respect to context
switch, otherwise overhead is too high.
• The average waiting time of RR is often long, but
• The average response time is often less, good for multi/user /processing.
Dr. Tarek Helmy, ICS-KFUPM 64
 Example of Round Robin Policy

Process Burst-time
P1 24
P2 3 Set quantum at 4 milliseconds
P3 3

P1 P2 P3 P1 P1 P1 P1 P1

0 4 7 10 14 18 22 26 30

• In the RR policy, the scheduling decision is made when:

– A process is done
– Expiration of the quantum time

• RR causes higher average turnaround time than SJF,

• But improves the average response time and fairness (starvation free)

Dr. Tarek Helmy, ICS-KFUPM 65

 Smaller Time Quantum Increases Context Switches

Dr. Tarek Helmy, ICS-KFUPM 66

 RR Policy Example: Preemptive

Process # Arriv al Time Burst Length Priority

P1 0 6 1 Time quantum:
P2 3 15 1 3 units
P3 9 3 1
P4 14 4 1
P5 17 2 1

P1: 6

P1

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 67

 RR Policy Example: Preemptive

Process # Arriv al Time Burst Length Priority

P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1 3 units
P4 14 4 1
P5 17 2 1
P1: 3
P1: 6 P2: 15

P1 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 68

 RR Policy Example: Preemptive

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1 3 units
P4 14 4 1
P5 17 2 1

P1: 3 P2: 12
P1: 6 P2: 15 P1: 3

P1 P2 P1

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 69

 RR Policy Example: Preemptive

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1 3 units
P4 14 4 1
P5 17 2 1

P1: 3 P2: 12 P3: 3

P1: 6 P2: 15 P1: 3 P2: 12

P1 P2 P1 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 70

 RR Policy Example: Preemptive
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1
P4 14 4 1 3 units
P5 17 2 1

P1: 3 P2: 12 P3: 3 P2: 9

P1: 6 P2: 15 P1: 3 P2: 12 P3: 3

P1 P2 P1 P2 P3

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 71

 RR Policy Example: Preemptive

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1
P4 14 4 1 3 units
P5 17 2 1

P1: 3 P2: 12 P3: 3 P2: 9 P4: 4

P1: 6 P2: 15 P1: 3 P2: 12 P3: 3 P2: 9

P1 P2 P1 P2 P3 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 72

 RR Policy Example: Preemptive

Proces s # Arriv al Time Burs t Length Priority

P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1
P4 14 4 1 3 units
P5 17 2 1

P2: 6
P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2
P1: 6 P2: 15 P1: 3 P2: 12 P3: 3 P2: 9 P4: 4

P1 P2 P1 P2 P3 P2 P4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 73

 RR Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 9 3 1 Time quantum:
P4 14 4 1 3 units
P5 17 2 1

P2: 6 P4: 1
P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6
P1: 6 P2: 15 P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2

P1 P2 P1 P2 P3 P2 P4 P5

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 74

 RR Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1
P4 14 4 1 3 units
P5 17 2 1

P2: 6 P4: 1
P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6 P4: 1
P1: 6 P2: 15 P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6

P1 P2 P1 P2 P3 P2 P4 P5 P2

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 75

 RR Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1
P4 14 4 1 3 units
P5 17 2 1
P2: 6 P4: 1
P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6 P4: 1 P2: 3
P1: 6 P2: 15 P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6 P4: 1

P
P1 P2 P1 P2 P3 P2 P4 P5 P2
4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 76

 RR Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1 Time quantum:
P3 9 3 1
P4 14 4 1 3 units
P5 17 2 1

P2: 6 P4: 1
P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6 P4: 1 P2: 3
P1: 6 P2: 15 P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6 P4: 1 P2: 3

P
P1 P2 P1 P2 P3 P2 P4 P5 P2 P2
4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 77

 RR Policy Example
Proces s # Arriv al Time Burs t Length Priority
P1 0 6 1
P2 3 15 1
P3 9 3 1 Time quantum:
P4 14 4 1 3 units
P5 17 2 1

P2: 6 P4: 1
P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6 P4: 1 P2: 3
P1: 6 P2: 15 P1: 3 P2: 12 P3: 3 P2: 9 P4: 4 P5: 2 P2: 6 P4: 1 P2: 3

P
P1 P2 P1 P2 P3 P2 P4 P5 P2 P2
4

0 5 10 15 20 25 30

Dr. Tarek Helmy, ICS-KFUPM 78

 Round Robin Policy Characteristics

• It is a preemptive and can be non-preemptive. How?

• Used in time sharing environments, WHY?
• Sensitive to the length of the time quantum [Research area!] due to
the overhead of context switch.
• How to predict the best unit of CPU time to be given to the
processes?
• It favors CPU-bound processes than I/O-bound processes.
• I/O-bound processes are not favored [Research area!]
– Every time an I/O request occurs the process has to go back to
the end of the ready queue !!!
– Variation:
• Using a separate queue with higher priorities for processes
“returning” from I/O activities.

Dr. Tarek Helmy, ICS-KFUPM 79

 Round Robin Policy: Disadvantages

• It favors CPU-bound processes:

– An I/O bound process uses the CPU for a time that is less than the time
quantum then blocked waiting for I/O and latter added to the end of the
ready queue.
– A CPU-bound process runs for all its time quantum and is put back into
the ready queue (thus getting in front of blocked processes).
• A solution: use virtual Ready Queue:
– When an I/O-bound process has completed, the blocked process is
moved to an auxiliary queue which gets preference over the main ready
queue.
– A process dispatched from the auxiliary queue should be favored. i.e.
• Given more time quantum or more frequency. i.e.
– Take two processes from the auxiliary queue and one from the
ready queue.

Dr. Tarek Helmy, ICS-KFUPM 80

 Queuing for Virtual Round Robin

If the QT for processes in the RQ is T1

i.e. the QT for processes in the AQ is T2+ (T1-T3)

i.e. if the I/O process took T3 in the first turn

Dr. Tarek Helmy, ICS-KFUPM 81

 RR Policy and Fairness

• Tries to give a fair share of CPU to each process (i.e. 1/N of the CPU time to
each process if there are N processes of the same type).
• The algorithm should compute the ratio of actual CPU time used to the CPU
time that the process should have been assigned (for a fair CPU assignment)
– Runs the process with the lowest ratio until the ratio moves above its closest competitor.

• Takes into account the owner of the process and the number of processes
per user when scheduling in multiuser system.
• Is CPU time shared among the users or processes fairly?
– Example: User 1 has 4 processes, A, B, C, D, and user 2 has 1 process
E. Then with round robin scheduling the execution sequence would be: A
E B E C E D E …..

Dr. Tarek Helmy, ICS-KFUPM 82

 Priority Scheduling Policy

• A priority number is associated with each process (based on a

certain criteria).
• The CPU is allocated to the process with the highest priority (smallest
integer  highest priority)
– It can be Preemptive or Non-preemptive
– Equal priority processes are scheduled in FCFS order.
– SJF is a priority scheduling where the priority is assigned based
on the predicted next CPU burst time.
• Problem  Starvation (low priority processes may never execute)
– Aging: Increase the process priority the longer it stays ready but
is not run.

Dr. Tarek Helmy, ICS-KFUPM 83

 Example of Priority Scheduling: Non-Preemptive Version

Process Burst Time Priority

P1 10 3
P2 1 1
P3 2 4
P4 1 5
P5 5 2

P2 P5 P1 P3 P4

0 1 6 16 18 19

AWT= (TStart to use CPU – T Arrive )/n= (6+0+16+18+1)/5 = 8.2

Dr. Tarek Helmy, ICS-KFUPM 84

 Priorities Assignment

• Rate Monotonic Scheduling (RMS)

– A static/fixed priority assignment scheme
• Assign priorities inversely proportional to the execution times
• Shortest execution time first
• Longest execution time first
– According to recourses utilization
• A process with less recourses utilization first
• A process with more recourses utilization first
• Lower-priority may suffer from starvation.
• Solution: allow a process to change its priority based on its age or execution history.

• Earliest Deadline First (EDF)Scheduling (for real time processes)

– Dynamic priority assignment scheme.
– Priorities are assigned according to absolute deadlines:
• The earlier the absolute deadline, the higher the priority.
– EDF can achieve 100% CPU utilization while still guaranteeing all the deadlines.
• Dynamic Priority algorithms provide better processor utilization than Fixed Priority
algorithms.

Dr. Tarek Helmy, ICS-KFUPM 85

 Hybrid Priorities Assignment

• To improve predictability for critical processes, use a combination of

fixed and dynamic priority algorithms.

• Processes divided based on criticality into “critical and non-critical”

processes.

• Critical processes scheduled using fixed priority assignment.

• Non-critical processes scheduled based on dynamic priority

assignment.

Dr. Tarek Helmy, ICS-KFUPM 86

 Preemptive Priority Algorithm

• When a process arrives at the ready queue with a higher priority

than the running process, the new one preempts the running
process.

– Low priority processes can starve

– i.e. UNIX uses aging to prevent starvation

– aging: if a process has not received service for a long time, its
priority is increased again.

Dr. Tarek Helmy, ICS-KFUPM 87

 Multi-level Queue Scheduling
• Till now we were using one ready queue for all ready processes and one scheduling
algorithm to schedule all of them.
• A multi-level queue-scheduling (MLQ) algorithm partitions the ready queue into several
separate queues.
• Created for situations in which processes are easily classified into groups:
– Foreground (interactive) processes, Background (batch) processes.
• These two types of processes have different response-time requirements, and thus,
different scheduling polices are needed.
• The processes are permanently assigned to one queue, based on some properties of
the process. (e.g. memory size, priority, or type).
• Each queue has its own scheduling algorithm. For e.g. the foreground queue might be
scheduled by an RR algorithm (improve throughput under light load), while the
background queue is scheduled by a FCFS algorithm.
• Scheduling must be done between the queues.
– Fixed priority scheduling: Serve all from higher-priority then from lower-priority;
possibility of starvation.
– Time slice: Each queue gets a certain amount of CPU time which it can schedule
amongst its processes; e.g., 80% to foreground in RR 20% to background in FCFS
Dr. Tarek Helmy, ICS-KFUPM 88
 Multilevel Queue Scheduling

Example: A MLQ with 5 queues:

Highest priority
System Processes

Real Time Processes

Interactive Processes

Background Processes

Batch Processes
Lowest priority
Possibility I
• If each queue has absolute priority over lower-priority queues then no process in the lower priority
queues could run unless the processes of the highest-priority queue were all done.
• For example, in the above figure no process in the batch queue could run unless the queues for
system processes, interactive processes, and interactive editing processes will all empty.
Possibility II
• If there is a time slice between the queues then each queue gets a certain amount of CPU times,
which can then schedule it among the processes in its queue. For instance;
– 30% of the CPU time to System Processes using Priority Scheduling Policy
– 25 % of CPU time for RT processes using RR.
– 20 % of CPU time for Interactive using SRT
– 15% of the CPU time to background queue using SRTF.
– 10% of the CPU time to Batch processes using FCFS

Dr. Tarek Helmy, ICS-KFUPM 89

 Multilevel Feedback Queues

• The problems with Multilevel Queue Scheduling are that:

• One queue may be full while the other may be empty: leads to wasting CPU
time assigned to it.
• This could create some problems with CPU hogging and starvation.
• If a very long process is put on Q1, it could hog all of the time given to
Queue 1, preventing the other processes from running.
• Once a process is put on a queue, it must remain there till its competition.

• Proposed Solution
– Feedback queues attempt to solve this by allowing a process to move
between various queues by:
• Separating processes with different CPU-burst characteristics.
• Leaving I/O-bound and interactive processes in the higher-priority
queues.
• Migrating a process waiting too long in a lower-priority queue to a
higher-priority queue (aging can be implemented this way).

Dr. Tarek Helmy, ICS-KFUPM 90

 Multilevel Feedback Queues

Highest priority

Quantum = 2

Quantum = 4

Quantum = 8

FCFS
???

Lowest priority

Dr. Tarek Helmy, ICS-KFUPM 91

 Multilevel Feedback Queues

Highest priority
Quantum = 2

Quantum = 4

Quantum = 8

FCFS

Lowest priority

Dr. Tarek Helmy, ICS-KFUPM 92

 Multilevel Feedback Queues

• Multilevel-feedback-queue scheduler implementation affected by the

following parameters:
– Number of queues
– Scheduling algorithm for each queue
– Method used to determine when to upgrade a process.
– Method used to determine which queue a process will enter
when that process needs service (i.e. all ready processes enter
at the end of queue 0) .
– Method used to determine when to downgrade a process to a
lower-priority queue (i.e. move down one level at the end of each
quantum)

Dr. Tarek Helmy, ICS-KFUPM 93

 Ch: 5 Process Scheduling
• Schedulers overview (i.e. long, short, medium term schedulers)
• Scheduler, Dispatcher and Swapper modules in the OS
• Types of Scheduling Algorithms
– Preemptive, and Non-preemptive
• Which of the Process states transition are Preemptive and which are Non-preemptive.
• Categories of Processes (i.e. batch, interactive, real-time) and the appropriate scheduling
• Evaluation Criteria of Scheduling Algorithms (Fairness, Throughput, Resources' utilization,
WT, TT, RT, Context Switches, complexity)
• Ideal Process Scheduling Algorithm (i.e. should maximize what and minimize what)
• Factors affect the scheduler decision & implementation (i.e. Preemption frequency, type of the
process, resolving conflicting requests, Page fault frequency, etc. )
• Scheduling Algorithms
– First-Come, First-Served (FCFS)
– Shortest Job First (SJF)
– Shortest Remaining Time First (SRTF)
– Process Burst Length Prediction
– Round Robin, its Characteristics and Disadvantages
– Priority-Based: Priorities assignment, its disadvantages and how to cover them
– Multilevel Queue and Multilevel Feedback Queue
– Example of MLFQ Scheduling algorithm
• Scheduling Algorithms Evaluation
– Deterministic Model, Queuing Model, Simulations, Implementation
• Scheduling Polices in Different OS
• Scheduling in Multiple-Processor Systems
• Introduce Ch. 6 which is going to be studied by yourself
Dr. Tarek Helmy, ICS-KFUPM 94
 Example: Multilevel Feedback Queue

• Three queues:
– Q0 – Time quantum 8 milliseconds
– Q1 – Time quantum 16 milliseconds
– Q2 – Time quantum , what is the appropriate scheduling
algorithm?
• Scheduling
– A new job enters queue Q0, when it gains CPU, it receives 8
milliseconds.
– If it does not finish in 8 milliseconds, it moves to queue Q1.
– At Q1, the job receives 16 milliseconds, if it still does not
complete, it is preempted and moved to queue Q2.

Dr. Tarek Helmy, ICS-KFUPM 95

 Multilevel Feedback Queues

Q0 = 8

Q1 = 16

Q2 = FCFS

Dr. Tarek Helmy, ICS-KFUPM 96

 Multilevel Feedback Queues

P1 P2 P3 P4 P1 P2 P4 P1 P4
0 8 16 21 29 45 47 63 67 69
• In Q0 (Quantum time = 8): P BT
– P1 uses 8 and needs 20 more P1 28
– P2 uses 8 and needs 2 more P2 10
– P3 uses 5 and terminates P3 5
– P4 uses 8 and needs 18 more P4 26

• So P1, P2, P4 will go to Q1 P BT

(Quantum time =16): P1 20
P2 2
P4 18

Dr. Tarek Helmy, ICS-KFUPM 97

 Multilevel Feedback Queues

P1 P2 P3 P4 P1 P2 P4 P1 P4
0 8 16 21 29 45 47 63 67 69
• P1, P2, P4 in Q1 (Quantum time =16): P BT
– P1 uses 16 and needs 4 more; P1 20
– P2 uses 2 and terminates;
P2 2
– P4 uses 16 and needs 2 more;
P4 18
• So P1(4), P4(2) will go to Q2 (Quantum
time = )
– P1 uses 4 and terminates;
– P4 uses 2 and terminates.
P BT
• AWT, ART & ATT ?
P1 4
• You need to calculate them for all
processes. P4 2

Dr. Tarek Helmy, ICS-KFUPM 98

 Comparison of the Algorithms

• After studying the following scheduling algorithms:

1. First-Come, First-Served (FCFS)
2. Shortest Job First (SJF)
3. Shortest Remaining Time First (SRTF)
4. Round robin
5. Priority Scheduling
6. Multilevel queue
7. Multilevel feedback queue
• Which one is best?
• The answer depends on:
– The system workload (which is extremely variable),
– The type of concurrently running processes (batch, real time, interactive,
etc.),
– Hardware support for the dispatcher,
– Relative weighting of the performance metric criteria (response time, CPU
utilization, throughput, etc.)
• Hence the answer depends on too many factors.

Dr. Tarek Helmy, ICS-KFUPM 99

 CPU Scheduling Algorithm Evaluation

• Define the criteria for evaluation and comparison:

– Maximize CPU utilization with maximum response time < threshold
value.
• Keep the CPU as busy as possible.
– Maximize throughput, how?
• Make the CPU executes and completes as many processes as it
can.
• Environment in which the scheduling algorithm is used will change
– An algorithm may be good today, but not good tomorrow?

• Evaluation methods
– Deterministic modeling
– Queuing models
– Simulations
– Implementation

Dr. Tarek Helmy, ICS-KFUPM 100

 Deterministic Modeling

• Deterministic modeling:

– Deterministic modeling is an analytic evaluation,

– Consider a predefined workload (set of processes) and evaluate

the performance of each algorithm for that workload to determine

the best decision is what.

– Simple, fast, and gives exact numbers,

– Too specific, and requires too much exact knowledge to be

useful.

Dr. Tarek Helmy, ICS-KFUPM 101

 Example: Compare FCFS, SJF, RR, &MLFQ

• Consider the following scenario with 3 processes P1, P2, and P3.

Processes CPU Time Arrival Time

P1 44 0
P2 9 1
P3 13 2

• Assume the context switch takes one time unit.

• Assume the scheduler itself takes 2 time units and is ready to act at
time 2.
• Let us say our goal is to maximize the throughput.
• To maximize the throughput,
– What is the evaluation Criteria we should calculate?
• Turn around time
• What is the best scheduling algorithm for executing them?
Dr. Tarek Helmy, ICS-KFUPM 102
 FCFS and SJF Polices

P1 P2 P3
2 46 56 70

ATT = [(46-0) + (56-1) + (70-2)] / 3 = 56.33

SJF
P2 P3 P1
2 11 25 70

ATT = [(70-0) + (11-1) + (25-2)] / 3 = 34.33

Processes CPU Time Arrival Time

P1 44 0
P2 9 1
P3 13 2

Dr. Tarek Helmy, ICS-KFUPM 103

 RR (Quantum = 10)

P1 P2 P3 P1 P3 P1 P1 P1
2 12 22 33 44 48 59 70 75

ATT = [(75-0) + (22-1) + (48-2)] / 3 = 47.33

Processes CPU Time Arrival Time

P1 44 0
P2 9 1
P3 13 2

Dr. Tarek Helmy, ICS-KFUPM 104

 MLFQ

Arrival Terminate
Q1: D1=10
Terminate
Q2: D2=20

Terminate
Q3: D3= infinity
P1 P2 P3 P1 P3 P1
2 12 22 33 54 58 73

ATT = [(73-0) + (22-1) + (54-2)] / 3 = 48.67

Processes CPU Time Arrival Time
P1 44 0
P2 9 1
P3 13 2
Dr. Tarek Helmy, ICS-KFUPM 105
 Compare FCFS, SJF, RR, &MLFQ

Scheduling Algorithm ATT Order

FCFS 56.33 4
SJF 34.33 1
RR 47.33 2
MLFQ 48.67 3

• That means, we can use case based learning mechanism and if

we have similar set of processes and our objective is to maximize
the throughput, the SJF is the best decision

Dr. Tarek Helmy, ICS-KFUPM 106

 Queuing Models

– Mathematical and statistical analysis

• Use the distribution of processes (CPU, I/O bound) arrival and the average
waiting time to compute the average length of ready queue which will affect
the (throughput, utilization, …).
• Example: Little’s formula
n=XW
n: average queue length (efficiency of the algorithm)
: average arrival rate (incoming processes)
W: average waiting time (depends on the scheduling algorithm)
n
 
• if =7 processes/sec, and W=2 sec.  n=14
• As far as the algorithm is minimizing the average queue length n, it will be
good.
• Can only handle simple algorithms and distributions.

Dr. Tarek Helmy, ICS-KFUPM 107

 Simulations

• Simulate a computer system model:

– Use data structure techniques to model queues, CPU, devices, timers, etc…
– Simulator modifies the system state to reflect the activities of the devices, CPU, the
scheduler, etc.
• Data to drive the simulation
– Random-number generated according to probability distributions
• Processes, CPU- and I/O-burst times, arrivals/departures
• Disadvantage – expensive
• Many open source CPU scheduling simulators are available on the web: i.e.
– http://cpuss.codeplex.com/
– http://www.jimweller.net/jim/java_proc_sched/applet.html
– http://www.clusterresources.com/products/maui-cluster-scheduler.php
– http://www.redbooks.ibm.com/abstracts/sg246038.html
– http://www.platform.com/workload-management/high-performance-computing
– http://en.wikipedia.org/wiki/Portable_Batch_System
– http://www.oracle.com/technetwork/oem/grid-engine-166852.html.
– http://oar.imag.fr

Dr. Tarek Helmy, ICS-KFUPM 108

 Implementation

• In an open source OS like UNIX, Linux

• Propose or refine the code of any scheduling algorithm,

• Put the coded algorithm in the real system for evaluation under real
operating conditions.

• Evaluate its performance empirically

• Difficulty

– Costly in coding the algorithm,

– Needs to have an open source OS.

Dr. Tarek Helmy, ICS-KFUPM 109

 Windows 2000, NT, XP, 7 & 8 Scheduling

• Windows uses a priority-based pre-emptive scheduling algorithm and 31

priority levels (31 is the highest).
• Priorities are assigned to both processes and threads.
• A running thread will run until it is preempted by a higher-priority one,
terminates, time quantum ends, calls a blocking system call
• Ranges of priorities are defined for processes and threads:
– Processes - idle, normal, and very high
– Threads - idle, below normal, normal, above normal, highest
• Dynamic priority may be adjusted up or down by the OS to reflect changing
conditions.
– Lower (not below base priority) when its time quantum runs out
– Priority raised up when it is released from a wait operation
• The higher level depends on the reason for wait
• Waiting for keyboard I/O gets a large priority increase
• Waiting for disk I/O gets a moderate priority increase
– Process in the foreground window get a higher priority

Dr. Tarek Helmy, ICS-KFUPM 110

 Windows Process Priorities

priority classes

levels
within
class

• Compute until end of slice: priority lowered

• After waiting for I/O: priority raised
– If I/O was mouse, keyboard: priority raised even more
– Favors interactive processes
• Foreground process (window) receives larger slice than background process (3x)

Dr. Tarek Helmy, ICS-KFUPM 111

 Windows 10 Scheduling

• Windows 10 uses a round-robin technique with a multi-level

feedback queue for priority scheduling.
• The ready queue is partitioned into multiple queues of different
priorities.
• The system assigns processes to queue based on their CPU burst
characteristic.
• If a process consumes too much CPU time, it is placed into a lower
priority queue.
• Process that waits too long in a lower priority queue may be moved
to a higher priority queue.

Dr. Tarek Helmy, ICS-KFUPM 112

 Linux Scheduling

• Separate Time-sharing and real-time scheduling algorithms

• Allow only processes in user mode to be preempted
– A process may not be preempted while it is running in kernel mode, even if a real-
time process with a higher priority is available to run
– Soft real-time system
• Two algorithms: time-sharing and real-time
• Time-sharing
– Prioritized credit-based – process with most credits is scheduled next
– 1 credit subtracted when timer interrupt occurs
– When credit = 0, process suspended, another one chosen
– When all ready processes have credit = 0, re-crediting occurs
• credits = credits / 2 + priority (internally: high priority = high int)
• So, if process did I/O before credits = 0, it has more credits next time
• I/O-bound, interactive processes accumulate more credits Real-time
scheduling
– Two real-time scheduling classes: FCFS (non-pre-emptive) and RR (pre-emptive)
• PLUS a priority for each process
– Always runs the process with the highest priority
• Equal priority  runs the process that has been waiting longest

Dr. Tarek Helmy, ICS-KFUPM 113

 Unix System Scheduling

• Adopts multilevel feedback with RR

• There is priority involved with each queue.
• If a running process does not block or complete, it is preempted within 1
second.
• Priority is determined according to process type and execution history.
Pj(i) = Basej + CPUj (i-1) + nicej
CPUj(i) = Uj(i) / 2 + CPUj (i-1) / 2
• Pj(i) = Priority of Pj at the beginning of interval i (lower values equal higher
priority).
• Basej = Base priority of Pj
• Uj(i) = CPU use of Pj in interval i
• CPUj(i) = Exponentially weighted average CPU use of Pj through interval i
• nicej =User control adjustable factor.

Dr. Tarek Helmy, ICS-KFUPM 114

 Unix System Scheduling

• The priority of a process is computed once per second.

• The purpose of the base priority is to divide all processes evenly

into fixed bands of priority levels.

• The CPU and nice components are restricted to prevent a

process from migrating out of its assigned band.

• The bands are swapper, block I/O device control, file

manipulation, character I/O device control and user processes.

Dr. Tarek Helmy, ICS-KFUPM 115

 Solaris 2 Scheduling

• Priority-based process scheduling

– Classes: real time, system, time sharing, interactive
• Each class has different priority and scheduling
algorithm
• Each thread assigns a scheduling class and priority
• Real-time processes run before a process in any other
class.
• System class is reserved for kernel use (paging,
scheduler)
– The scheduling policy for the system class does not
time-slice
• Time-sharing/interactive: multilevel feedback queue
• The selected thread runs on the CPU until it blocks, uses
its time slices, or is preempted by a higher-priority thread.
– Multiple threads have the same priority  RR

Dr. Tarek Helmy, ICS-KFUPM 116

 Summary

• CPU scheduling is the task of selecting a waiting process from the

ready queue and allocating the CPU to it. The CPU is allocated to
the selected process by the dispatcher.
• FCFS is the simplest scheduling algorithm but it can cause short
processes to wait very long.
• SJF provides the shortest average waiting time. Implementing SJF
is difficult, due to the difficulty in predicting the length of the next
CPU burst.
• SJF is a special case of the general priority scheduling algorithm,
which allocates the CPU to the highest-priority process. Both SJF
and priority may suffer from starvation. Aging is a technique to
prevent starvation.

Dr. Tarek Helmy, ICS-KFUPM 117

 Summary

• RR scheduling is appropriate for time-sharing systems. RR allocates

the CPU to the first process in the ready queue for q time units. The
major problem is the selection of the time quantum. If the quantum is
too large, RR degenerates to FCFS; if q is too small, overhead due to
context switching becomes excessive.
• The FCFS algorithm is non-preemptive
• The RR algorithm is preemptive.
• The SJF and priority algorithms may be either preemptive or non-
preemptive.
• Multi-level queue algorithm allows different algorithms to be used for
various classes of processes. The most common is a foreground
interactive queue which uses RR scheduling, and a background
batch queue which uses FCFS scheduling.
• Multi-level feedback queues allow processes to move from one
queue to another.

Dr. Tarek Helmy, ICS-KFUPM 118

 Multiple-Processor Systems: FYI only

• So far we looked at CPU scheduling algorithms Multi-Processor

for single processor systems. If multiple CPUs
exist, the scheduling problem becomes more
complex. SIMD MIMD
Classifications of Multiprocessor Systems:
• Loosely coupled multiprocessor, or clusters
– Each processor has its own memory and Shared Memory Distributed Memory
I/O channels. (tightly coupled) (loosely coupled)
• Functionally specialized processors
– Such as I/O processor
– Controlled by a master processor
• Symmetric Multi-Processors (SMP):
Master/Slave Symmetric Clusters
– The OS code and data structures are global Asymmetric (SMP)
(stored in the shared memory and equally CPU CPU
available to each processor).
– Controlled by operating system CPU CPU
Shared
– One copy of OS in memory, any CPU can Memory
CPU CPU
use it.
– OS must ensure the consistency of shared
CPU CPU
data.
Dr. Tarek Helmy, ICS-KFUPM 119
 Multiple-Processor Scheduling: Design Issues

• Scheduling on a multiprocessor involves 3 interrelated design issues:

– Assignment of processes to processors

• One of the design issues that must be decided is (which process to

assign to each processor?).

– Actual dispatching of a process [Process Scheduling]

– Use of multithreading on individual processors [Threads Scheduling]

• It is two dimensional problem

– Which processor runs the next process? (processor scheduler,

coordinator processor)

– Which process runs next? (process scheduler, per processor)

Dr. Tarek Helmy, ICS-KFUPM 120

 Assignment of processes to processors
1. Master/slave architecture (Asymmetric MP Scheduling)
• Main kernel functions always run on a particular
processor/master. Master is responsible for scheduling
• Slave processors execute user processes.
• Only one processor accesses the system data structures,
alleviating the need for data sharing.
• No coherency problems
• Failure of master brings down the whole system.
• Master can become a performance bottleneck.
2. Peer to Peer architecture (Symmetric MP Scheduling)
• OS executes on any processor
• Each processor does self-scheduling
• Processes may be in a global ready queue, or each
processor may have its own ready queue
• Mutual exclusion problems (Make sure two processors do
not choose the same process; needs lots of
synchronization)
• SMP is supported by all modern operating systems:
Windows, Solaris, Linux, Mac OS X.
Dr. Tarek Helmy, ICS-KFUPM 121
 Assignment of processes to processors

• Dynamic vs. static assignment of processes to processors: Types of policies:

• Static: decisions have been known to system
• Dynamic: uses load information
• Adaptive: policy varies according to load
• Exploit cache affinity: try to schedule a process/thread on the same processor
that a process/thread executed last.
• Context switch overhead
– Quantum sizes larger on multiprocessors than uni-processors
– Preemptive versus non-preemptive
• Measure of load: Queue lengths at CPU, CPU utilization
– Transfer policy: when to transfer/migrate a process?
• Threshold-based policies are common and easy
– Selection policy: which process to transfer/migrate?
• Prefer new processes
• Transfer/migration cost should be small compared to execution cost
– Select processes with long execution times
– Location policy: where to transfer/migrate the process?

Dr. Tarek Helmy, ICS-KFUPM 122

 Dynamic vs. Static Assignment

• The assignment of processes to processors can be static (till the process

completes) or dynamic (may change each time a process is dispatched).
• Static assignment maintains a separate ready queue for each processor.
– Loads can become unbalanced if the processes in one queue run longer
than those in another queue.
– Coordinating agent as a middle layer is needed to make assignment
decisions [Hot Research area!!].
– If processes on different processors are dependent, the schedulers on
the processors must synchronize the processes according to some
synchronization and resource access-control protocol.

• Dynamic assignment: use a Global queue and processes are allowed to

migrate from processor to processor.
– When a process returns to the Ready state, it returns to a shared queue
and when it is dispatched will run on the next free processor.

Dr. Tarek Helmy, ICS-KFUPM 123

 Multiprocessor Scheduling: Global Queue

• A single scheduling algorithm is used to schedule all tasks.

Important Difference:

– Schedule to any available processor. P1

P2
– May cause overheads if a processes
P3
migrates to a new processor.
P4
– Static scheduling reduces this problem.

• Appropriate only on tightly-coupled/multi-core

systems.

Dr. Tarek Helmy, ICS-KFUPM 124

 Multiprocessor Scheduling: Partitioning

• Partition processes so that each process always runs on the same

processor.
1. Dedicate short-term queue for each processor
2. Schedule processes on each processor using uni-processor algorithms.
3. Less overhead
Problems with Partitioning
P1
• Variations in computation costs may necessitate a re-
partitioning. Processor could be idle while another processor P2
has a backlog.
P3
• Which is good? Scheduling dependent processes or threads at
the same time across multiple CPUs or to the same CPU? P4
• Shall partitions size be fixed or dynamically modified?
• Equal-partition: constant number of processes allocated evenly
to all processors: Low overhead.
• Dynamic: dynamically reallocates processes to maximize
utilization: High utilization.
• Problematic because:
– Re-partitioning may cause task migrations…
• Partitioning is designed to avoid tasks migrations.

Dr. Tarek Helmy, ICS-KFUPM 125

 Multiprocessor Scheduling: Processor Affinity

• On a SMP, each processor has its own local cache or caches

• Scheduling a process where it has “Cash affinity” improves performance by reducing
cache penalties.
• Suppose process P is running on CPU # 0 for this time slice
• The cache of CPU # 0 contains many entries copied from P’s address space (PCB)
• After context switch, P is assigned to CPU # 1
• The cache of CPU # 1 has no entries of P’s address space
• Must re-fetch from main memory
– Very expensive
• If P was assigned to CPU # 0 instead, re-fetching would be minimized
• Some OSs try to re-assign a process to the CPU it was running last. This is called
Processor/Cash affinity
• Cash affinity can be either soft or hard affinity
– Soft affinity: “best effort” to get same CPU but no guarantee
– Hard affinity: A process can request and get same CPU, guaranteed
• Load Balancing:
– SMP systems must keep the workload balanced across all processors
– Only necessary in systems where each processor has its own queue.

Dr. Tarek Helmy, ICS-KFUPM 126

 Multiple-Processor Scheduling: Load Balance

• Two general approaches

• Push migration: A specific process periodically checks the load on each
processor.
– If it finds an imbalance, it moves (pushes) processes to idle processors
• Pull migration: An idle processor pulls a waiting process from a busy
processor.
• Hybrid. Uses both push and pull.
– Example: Linux scheduler implements both.
– Linux runs balancing algorithm every 200 milliseconds (push)
– Or whenever the run queue for a processor is empty (pull)
• Problem: load balancing often counteracts the benefits of processor affinity
– If using push or pull migration will take a process from its processor
– This violates processor affinity
– No absolute rule governing which policy is best
• In some systems an idle processor always pulls a process from a
non-idle process.
• In other systems process are moved only if the imbalance exceeds a
threshold.

Dr. Tarek Helmy, ICS-KFUPM 127

 Ch. 6 Process Synchronization
• To practice collaborative learning as we agreed, to foster self learning skill, and to help us
progress in the course,
• You need to study Ch. 6 “Process Synchronization” and it will be included in the Major Exam
II.
• Office hours can be used to answer any inquires about this chapter.
• Your objectives out of this chapter is to know:
– Why synchronization? Necessity of synchronization in OS
– How do processes work with resources that must be shared between them?
– What is a critical section?
– How to ensure that only one process can access the critical section?
– What is atomic operation? It executes without interruptions, all or none.
– Dangers of handling the critical section without synchronization.
– Different algorithms to synchronize two processes enter of critical section.
• Evaluating synchronization algorithms of handling a critical section where every algorithm
should allow Mutual Exclusion, Progress and bounded waiting.
– Synchronization tools
– Semaphores, and types of Semaphores
– Incorrect usages of Semaphores
– Classical problems of synchronization
– Monitors
– Synchronization methods in different OSs.
Dr. Tarek Helmy, ICS-KFUPM 128
 The End!!

Thank you

Any Questions?

Dr. Tarek Helmy, ICS-KFUPM