Lecture 7 Process Management and Single Processor Scheduling

Outline

1. Concept of Processor scheduling

• Timing and strategy for processor scheduling
• Criteria for comparing scheduling algorithms

2. Scheduling Algorithms

Time-Division Multiplexing of CPU resources

• Process switching: switching the current occupants of CPU resources

  • Save the execution context (CPU status) of the current process in the PCB
  • Restore the execution context of the next process

• Processor scheduling

  • Select the next process that will occupy the CPU from the ready queue
  • Select an available CPU to the ready process from multiple available CPUs

• Scheduler: a kernel function that picks a ready process to run

• Scheduling Algorithm

  • What principles are used to select a ready process?

Scheduling timing

• The conditions when the kernel starts to schedule
  • Process switches from running state to waiting/ready state
  • The process was terminated
• Non-preemptive systems
  • Current process voluntarily yields CPU
• Preemptive system
  • Interrupt requests are handled by the interrupt handler

Scheduling Algorithm

Determine how to select the next process from the ready queue for execution

• Issues to be addressed:
  • Based on what criteria is selection made?
  • Which process in the ready queue is picked?
• Scheduling Algorithms
  • Scheduling strategies implemented in kernel
• Criteria for comparing scheduling algorithms
  • Which strategy/algorithm is better?

Usage Patterns of Processor Resource

• Process alternates between CPU computation and I/O operations

  • Each scheduling decision determines which task is assigned to CPU for the next computation.
  • With time slicing, a process may be forced to yield the CPU before completing its current computation.

  

Outline

1. Concept of Processor scheduling
   • Timing and strategy for processor scheduling

- Criteria for comparing scheduling algorithms

2. Scheduling Algorithms

Criteria for comparing scheduling algorithms

• CPU Utilization: the percentage of time when the CPU is busy
• Throughput: the number of processes completed per unit of time
• Turnaround time: the total time the process takes from initialization to completion (including waiting time)
• Ready-waiting time: the total time that the ready process waits in the ready queue
• Response time: the total time from submitting a request to producing a response
• Fairness: ensuring processes have equal access to resources, such as CPU time

Criteria for comparing Scheduling Algorithms: Throughput and Latency

• Requirements of scheduling algorithms: desire "faster" service
• What is "faster"?
  • High bandwidth for file transfers, high throughput for scheduling algorithms
  • Low latency when playing games, low response time for scheduling algorithms
  • These two factors affect each other
• Analogy with water pipes
  • Low latency: When you drink water, you want the water to flow out as soon as you open the tap
  • High bandwidth: You want a large volume of water to flow simultaneously from the faucet when filling a swimming pool, regardless of latency

Criteria for Comparing Scheduling Algorithms: Response Time

• Reduce response time
  • Handle the user's request promptly, and provide feedback to the user as soon as possible
• Reduce fluctuations of average response time
  • In interactive systems, predictable response time is more important than low average response time with high variance
• Low-latency scheduling improves user interaction experience
  • If the cursor on the screen does not move when the mouse is moved, the user may restart the computer
• Response time is the computing delay of the OS

Criteria for Comparing Scheduling Algorithms: Throughput

• Increase throughput
  • Reduce overhead (OS overhead, context switching)
  • Efficient utilization of system resources (CPU, I/O devices)
• Reduce waiting time in the ready queue
  • Reduce waiting time for each ready process
• The OS needs to ensure that throughput is not affected by user interaction
  • Even if there are many interactive tasks
• Throughput is the computing bandwidth of the OS

Criteria for Comparing Scheduling Algorithms: Fairness

If one user runs more processes than other users, is it fair? What can be done?

• Definition of fairness
  • Ensure each process has equal CPU time
  • Ensure each process has the same waiting time in the ready queue
• Fairness usually increases average response time

Outline

1. Concept of processor scheduling

2. Scheduling Algorithms

• First-come First-served(FCFS), Short job first(SJF)
• Shortest remaining time(SRT), Highest Response Ratio Next(HRRN)
• Round-Robin(RR)
• MultiQueue(MQ), Multi-Level Feedback Queue(MLFQ)
• Fair Share Scheduling(FSS)

First Come, First Served (FCFS)

FCFS: First Come, First Served

• Processes are scheduled in the order they arrive in the ready queue.
• When a process becomes waiting or end state, the next process in the ready queue occupies the CPU
• Metric
  • Turnaround time of FCFS

An Example of FCFS

• Example: 3 processes with calculation times of 12,3,3

Characteristics of FCFS

• Advantages: Simple
• Disadvantages:
  • High fluctuation in the average waiting time
  • Short jobs/tasks/processes may have to wait for a long one to finish.
  • Low utilization of I/O and CPU resources
    • CPU-intensive processes will lead to a situation that when I/O devices are idle, I/O-intensive processes have to wait.

Short Job First (SJF)

Short Job First

• Selects the process with the shortest expected CPU time from the ready queue to run.

• The ready queue is sorted by expected execution time.

  

Characteristics of SJF

SJF has the best average turnaround time.

Can changing the order of process execution reduce the average turnaround time?

Characteristics of SJF

• May cause starvation

  • A sequence of short jobs may prevent long jobs from obtaining CPU resources.

• Requires knowledge of the future

  • How to estimate the duration of the next CPU calculation?
  • Simple solution: ask the user
    • Kill the corresponding process if the user cheats
    • What to do if the user doesn't know?

Execution Time Estimation of SJF

• Use historical execution time to predict future execution time

$\tau_{n+1} = \alpha t_n+(1-\alpha) \tau_n, where 0\le \alpha \le 1$
$t_n$ -- the nth CPU calculation time
$\tau_{n+1}$ -- CPU calculation time estimate for the n+1th time

$\tau_{n+1} = \alpha t_n+(1-\alpha) \alpha t_{n-1} + (1-\alpha) (1-\alpha) \alpha t_{n-2} + .. .$

Execution Time Estimation of SJF

• Execution Time Estimation
  

Shortest Remaining Time (SRT)

Shortest Remaining Time (SRT)

• SRT supports preemption, i.e., if a new process is ready and its service time is smaller than the remaining time of the current process, then the new process will be scheduled to execute.

Highest Response Ratio Next, HRRN

Highest Response Ratio Next, HRRN

• HRRN is mainly used for job scheduling.
• This algorithm is a balanced combination of FCFS and SJF, considering both the waiting time and estimated running time of each job.
• In each job scheduling, calculate the response ratio of each job in the queue and select the job with the highest response ratio to run.

Highest Response Ratio Next, HRRN

• Select the process in the ready queue with the highest response ratio R value.

$R=(w+s)/s$
w: ready waiting time (waiting time)
s: execution time (service time)

• Improve the Shortest Job First algorithm.
• Focus on the waiting time of the process.
• Prevent indefinite postponement

Outline

1. Concept of processor scheduling
2. Scheduling algorithm

• First-come, First-served(FCFS), Short job first(SJF)
• Shortest remaining time(SRT), Highest Response Ratio Next(HRRN)

Round-Robin(RR)

• MultiQueue(MQ), Multi-Level Feedback Queue(MLFQ)
• Fair Share Scheduling(FSS)

Round-Robin (RR)

RR, Round-Robin

• Time slice
  • The basic time unit for allocating processor resources.
• Algorithm ideas
  • At the end of the time slice, switch to the next ready process according to FCFS algorithm
  • Execute a time slice q every (n – 1) time slices.
    

An example of RR

Time slice length of RR

• Overhead of the RR algorithm: additional context switching
• Too large time slice
  • Long waiting time, degenerates into FCFS in the extreme case.
• Too small time slice
  • Highly responsive, but produces lots of context switches
  • A lot of context switching's overhead affects system throughput
• Time slice length selection goal:
  • Choose an appropriate time slice length
  • Empirical rule: keep the context switching overhead within 1%

Comparison between FCFS and RR

Outline

1. The concept of processor scheduling
2. Scheduling algorithm

• First-come, First-served(FCFS), Short job first(SJF)
• Shortest remaining time(SRT), Highest Response Ratio Next(HRRN)
• Round-Robin(RR)

MultiQueue(MQ), Multi-Level Feedback Queue(MLFQ)

• Fair Share Scheduling(FSS)

MultiQueue Algorithm, MQ

• The ready queue is divided into multiple independent subqueues
  • For example, foreground process (interactive) sub-queue, background process (batch processing) sub-queue
  • Processes with the same priority belong to a certain queue and cannot cross queues
• Each queue has its own scheduling policy

  • Such as: foreground process – RR, background process – RR/FCFS with a large time slice

  • Rule 1: If A's priority > B's priority, run A (do not run B).

  • Rule 2: If A's priority = B's priority, run A and B alternately.

MultiQueue Algorithm, MQ

• Scheduling between queues
  • Fixed priority
    • Handle foreground (interactive) processes first, then process background processes
    • May cause starvation
  • Round-Robin
    • Each queue gets a fixed amount of CPU time to schedule its processes
    • For example: 80% CPU time is used for foreground processes, 20% CPU time is used for background processes

Multi-Level Feedback Queue, MLFQ

MLFQ, Multi-Level Feedback Queue

• In 1962, Prof. Corbato from MIT first proposed the multi-level feedback queue, which was applied to CTSS (Compatible Time-Sharing System)
• Solve two problems
  • How to optimize turnaround time without knowing how long the job will run
  • How to reduce the response time and provide a good interactive experience for users?

Multi-Level Feedback Queue, MLFQ

• Key issues: how to schedule without complete knowledge?
  • How to build a scheduler that reduces both response time and turnaround time when the length of the process is unknown?
• Inspiration: Learn from History
  • Predict the future with historical experience
• Inherit the scheduling rules of MultiQueue
  • If A's priority > B's priority, run A (do not run B)
  • If A's priority = B's priority, run A and B in a round-robin/FIFO manner

Multi-Level Feedback Queue, MLFQ

Basic scheduling rules

• When a job enters the system, it is placed at the highest priority (top-level queue)
• If the process doesn't finish within the current time slice, it will be moved to the next lower priority queue.
• If a job actively yields the CPU within its time slice, its priority remains the same
• Time slice size increases as priority level increases
  

An Example of MLFQ with three priority queues

• CPU-intensive processes enter the highest priority queue first;
• After executing a 1ms time slice, the scheduler will reduce the priority of the process by 1 and moves it to the next highest priority queue;
• After executing a 2ms time slice, the process enters the lowest priority queue of the system, stays there, and executes using a 4ms time slice.

Multi-Level Feedback Queue, MLFQ

• Characteristics of the MLFQ

  • The priority of CPU-intensive processes drops quickly
  • I/O-intensive processes stay at high priority

• Potential problems

  • CPU-intensive processes may starve
  • Malicious processes may try to stay at high priority
    

Multi-Level Feedback Queue Scheduling Algorithm MLFQ

Basic scheduling rules

• If A's priority > B's priority, run A (do not run B)
• If A's priority = B's priority, run A and B in a round-robin/FIFO manner
• When a job enters the system, it is placed at the highest priority (the top-level queue)
• Once a job has used up its time slice in a certain level (regardless of how many times it has voluntarily given up the CPU), its priority is reduced (moved to a lower-level queue)
• After a period of time S, move all jobs in the system to the highest priority queue

Fair Share Scheduling Algorithm, FSS

FSS, Fair Share Scheduling

• Control users' access to system resources

  • Different users have multiple processes
  • Resources are allocated according to user priority
  • Ensures that unimportant users cannot monopolize resources
  • Unused resources are distributed proportionally

Characteristics of different scheduling algorithm

• First Come First Served (FCFS) algorithm
  • Poor average waiting time
• Shortest job first(SJF) algorithm
  • Minimal average turnaround time
  • Requires accurate prediction of calculation time
  • No preemption allowed; may cause starvation
• Shortest Remaining Time (SRT) algorithm
  • Improve the short job first algorithm, allowing preemption
  • May cause starvation

Characteristics of different scheduling algorithm

• Highest Response Ratio Next (HRRN) algorithm
  • Based on SJF, non-preemptive
  • Consider both the waiting time and the estimated run time of each job
• Round Robin (RR) algorithm
  • Fair, but has a poor average waiting time
• Multi-Level Feedback Queue (MLFQ) algorithm
  • Integration of multiple algorithms
• Fair share scheduling(FSS) algorithm
  • Fairness is the primary consideration

Summary

1. Concept of Processor Scheduling

• Timing and strategy of processor scheduling, criteria for comparing scheduling algorithms

2. Scheduling Algorithms

• FCFS, SJF, SRT, HRRN
• RR
• MQ, MLFQ, FSS