Advanced Operating Systems

Advanced Scheduling Algorithms

Session 3

Yayati Gupta
[email protected]

Mahindra University

December 17, 2024

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 1 / 30
Advanced Operating Systems

Outline
1 Basic Scheduling algorithms: Recap
2 Multilevel Queue Scheduling
Basic Multilevel Queue Scheduling
Multilevel Feedback Queue Scheduling
3 Linux Fair Share Scheduling Algorithms
Lottery Scheduling
Completely Fair Scheduling
4 Multiprocessor Scheduling (MS)
Single Queue MS
Multi Queue MS
5 Real Time Scheduling Algorithms
Rate Monotonic Scheduling
Earliest Deadline First
Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 2 / 30
Basic Scheduling algorithms: Recap
Advanced Operating Systems
Basic Scheduling algorithms: Recap

Basic Scheduling Algorithms

In the following process state transition diagram for a uniprocessor system, assume that there
are always some processes in the ready state.
1 If a process makes a transition D, it would

result in another process making transition

A immediately.
2 A process P2 in blocked state can make
transition E while another process P1 is in
running state.
3 The OS uses preemptive scheduling.
4 The OS uses non-preemptive scheduling.

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 4 / 30
Advanced Operating Systems
Basic Scheduling algorithms: Recap

Non Preemptive Scheduling Algorithms

1 First Come First Served (FCFS): Ready Queue Addition: O(1), Scheduling: O(1)
2 Shortest Job First (SJF):
Unsorted: RQ Addition: O(1), Scheduling: O(n)
Sorted list: RQ Addition: O(n), Scheduling: O(1)
Min heap: RQ Addition: O(log n), Scheduling: O(log n)
Additional overhead of predicting burst time: O(1) in case of exponential averaging
3 NP Priority Scheduling: Similar to SJF.
User defined
System processes can be given higher priorities
Priority of I/O bound jobs > Priority of CPU bound jobs
Can be based on deadlines as in Real time scheduling

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 5 / 30
Advanced Operating Systems
Basic Scheduling algorithms: Recap

Preemptive Scheduling Algorithms

1 Round Robin Scheduling (RR)

2 Shortest Remaining Time First (SRTF)
3 P Priority Scheduling

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 6 / 30
Advanced Operating Systems
Basic Scheduling algorithms: Recap

Convoy Effect

P# AT CB IOB CB IOB CB IOB

P0 0 100 6 100 6 100 6
P1 1 1 6 1 6 1 6
P2 1 1 6 1 6 1 6
P3 1 1 6 1 6 1 6
P4 1 1 6 1 6 1 6
P5 1 1 6 1 6 1 6

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 7 / 30
Advanced Operating Systems
Basic Scheduling algorithms: Recap

Basic Scheduling Algorithms

Algorithm Pros Cons

FCFS Simple implementation, Fair Convoy effect
SJF/SRTF Optimal Needed prediction, Starvation
Priority Scheduling Flexible Starvation
Round Robin Fair Frequent context switches, Op-
timal time quantum
1 SJF/SRTF: Compute intensive tasks (Background processes)
2 Round Robin: I/O bound tasks (Foreground processes)
3 Priority: System processes
No one size fits all!

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 8 / 30
Multilevel Queue Scheduling
Advanced Operating Systems
Multilevel Queue Scheduling
Basic Multilevel Queue Scheduling

Multilevel Queue Scheduling (One size doesn’t fit all)

1 CPU bound
processes: SRTF
2 I/O bound
processes: RR
3 Partition the ready
queue into
multiple queues
with varying
priorities
4 Among queues:
Fixed Priority/
Timeslice

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 10 / 30
Advanced Operating Systems
Multilevel Queue Scheduling
Basic Multilevel Queue Scheduling

Multilevel Queue Scheduling Challenges

1 Starvation in low
priority queues
2 How to differenti-
ate/approximate?
CPU bound
processes
I/O bound
processes

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 11 / 30
Advanced Operating Systems
Multilevel Queue Scheduling
Multilevel Feedback Queue Scheduling

Most general and Turing Award winning Algorithm

1 Smartly infers the type of the processes (no history needed)
Qi : RR with time quantum 2i
Qn−1 : FCFS
2 Timeslice expired at Qj : Move to Qj+1
3 Windows: 32 priority levels/queues with top 50% for real-time tasks (e.g. multimedia,
gaming etc.)

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 12 / 30
Advanced Operating Systems
Multilevel Queue Scheduling
Multilevel Feedback Queue Scheduling

Question

Given:
Queues: Q1 (Time Quantum: 3), Q2 (Time Quantum: 6), Q3: FCFS
Processes:
Process Arrival Burst I/O
P1 0 6 4
P2 3 10 3
P3 6 4 2

Draw the Gantt chart and find waiting time and turnaround time.

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 13 / 30
Linux Fair Share Scheduling Algorithms
Advanced Operating Systems
Linux Fair Share Scheduling Algorithms
Lottery Scheduling

Linux Scheduling Algorithms

1 Designed for responsiveness fairness (avoid starvation)

Server oriented OS
HPC
Distributed OS
Cloud computing environments
2 Lottery Scheduling: Probabilistic Fairness
3 RR + Priority
4 A simple Python implementation
5 Variants: Tickets transfer, ticket inflation, hierarchical lottery

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 15 / 30
Advanced Operating Systems
Linux Fair Share Scheduling Algorithms
Lottery Scheduling

Question

Given:
Process P1: Priority = 5, Burst Time = 8 units.
Process P2: Priority = 3, Burst Time = 6 units.
Process P3: Priority = 4, Burst Time = 4 units.
Time Quantum (TQ) = 1 unit.
1 Calculate the probability of each process being selected for execution.
2 What is the expected CPU time that Process P2 will get in the first 4 units of
time?
3 What happens to the expected CPU time for P2 if the Time Quantum (TQ) is
changed to 2 units instead of 1 unit? Explain the difference in your answer.

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 16 / 30
Advanced Operating Systems
Linux Fair Share Scheduling Algorithms
Completely Fair Scheduling

Completely Fair Share Scheduling

1 Which process to run?: one with least Virtual Runtime (vruntime): proc/pid/sched
sched latency
2 How long to run?: n : /sys/kernel/debug/sched/latency ns
If n is very large, use min granularity: /sys/kernel/debug/sched/min granularity ns
3 Other parameters
Nice value: Determines priority
Each nice value maps to a weight: Kernel scheduler code(line 10,114)
How long to run: Adjust the formula based on wi .
A similar adjustment is made for vruntime.
4 Implemented with Red-Black trees for high performance.

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 17 / 30
Advanced Operating Systems
Linux Fair Share Scheduling Algorithms
Completely Fair Scheduling

Question

Consider the following processes with their corresponding weights and burst times:
Process 1: Weight = 1024, Burst Time = 12
Process 2: Weight = 512, Burst Time = 18
Process 3: Weight = 2048, Burst Time = 30
Given the following parameters for Completely Fair Share Scheduling (CFSS):
Latency = 24
Minimum Granularity = 3
Task: Find the CFSS schedule for the given processes.
Note: You can assume all units to be in milliseconds.

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 18 / 30
Multiprocessor Scheduling (MS)
Advanced Operating Systems
Multiprocessor Scheduling (MS)

Basics

Figure: Two CPUs with caches sharing memory

1 Multithreading with parallelism: Faster applications

2 Cache Coherence (Resolved by the hardware and semaphores)
3 Cache Affinity
Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 20 / 30
Advanced Operating Systems
Multiprocessor Scheduling (MS)
Single Queue MS

Single Queue MS

Simple to implement
The queue needs to be locked: Increases contention, reduces scalability
Cache affinity disregarded

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 21 / 30
Advanced Operating Systems
Multiprocessor Scheduling (MS)
Multi Queue MS

Multi Queue MS

No locking required
Cache affinity is preserved (Recall cpu core.c): CFSS uses MQMS.
Load imbalance problem
Solution: Use job migration

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 22 / 30
Real Time Scheduling Algorithms
Advanced Operating Systems
Real Time Scheduling Algorithms

Real Time Scheduling Algorithms

1 The goal is to maximise throughput, maximise response time, meet deadlines
2 Fairness doesn’t help you meet deadlines.

Figure: Cowboy Scheduling Algorithm

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 24 / 30
Advanced Operating Systems
Real Time Scheduling Algorithms

Cowboy Scheduling Algorithm

Priority scheduling (as used in Windows/Linux) can work for soft real time systems
Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 25 / 30
Advanced Operating Systems
Real Time Scheduling Algorithms

Timing Parameters for the process

1 Period, p (Rate = p1 )
2 Execution time, t
3 Deadline, d
4 0≤t≤d ≤p

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 26 / 30
Advanced Operating Systems
Real Time Scheduling Algorithms
Rate Monotonic Scheduling

Rate Monotonic Scheduling

1 Static priority scheduling
2 Priority ∝ Rate

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 27 / 30
Advanced Operating Systems
Real Time Scheduling Algorithms
Rate Monotonic Scheduling

Rate Monotonic Shceduling: Working Example

We have two processes P1 and P2. The periods for P1 and P2 are 50 and 100,
respectively-that is, Pl =50 and P2 = 100. The processing times are t1 = 20 for P1 and t2 =
35 for P2 . The deadline for each process requires that it complete its CPU burst by the start
of its next period.

Is it schedulable?
CPU utilisation ≤ 100%
CPU utilisation ≤ n(2n − 1)
= 82.8% for n = 2
= 69.3% for n → ∞

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 28 / 30
Advanced Operating Systems
Real Time Scheduling Algorithms
Earliest Deadline First

Earliest Deadline First

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 29 / 30
Advanced Operating Systems
Real Time Scheduling Algorithms
Earliest Deadline First

Thank You

Questions?

Yayati Gupta (Mahindra University) Advanced Operating Systems December 17, 2024 30 / 30