1dv012 CPU Scheduling - orion.lnu.se
52
Transcript of 1dv012 CPU Scheduling - orion.lnu.se
Reasonable model
Actual Conceptual Time line
Different views on multi programming
Process States
Scheduling
• Processes can be described as either:
• I/O-bound process – spends more time doing I/O than computations, many short CPU bursts
• CPU-bound process – spends more time doing computations; few very long CPU bursts
Scheduling
CPU burst
Preemption
• Major classification
• Preemptive vs non-preemptive scheduling
• Non-preemptive: scheduling under 1 and 4
• A process switches to a waiting state only as a function of its own behaviour, i.e. when it invokes OS services, or when it terminates
• Preemptive: all other
• A process can be “forced” to move to a waiting state
Preemption
• Non-preemptive scheduling is “easier”
• Consider access to shared data
• Consider preemption while in kernel mode
• Consider interrupts occurring during crucial OS activities
• Cost: Maintaining consistent system state while the processes are suspended in the midst of critical activity
Dispatcher• Dispatcher module gives control of the CPU to the
process selected by the short-term scheduler; this involves:
• switching context
• switching to user mode
• jumping to the proper location in the user program to restart that program
• Dispatch latency – time it takes for the dispatcher to stop one process and start another running
Scheduling Metrics: User
• Performance related
• response time: time it takes to produce the first response
• turnaround time: time spent from the time of “submission” to time of completion
• deadlines: the time within which the program must complete (the policy must maximize percentage of deadlines met)
Scheduling Metrics: User
• Other
• predictability: expectation that the job runs the same regardless of system load
Scheduling Metrics: System
• Performance related
• waiting time: time spent waiting to get the CPU
• throughput: the number of processes completed per unit time (directly affected by the waiting time)
• CPU utilization: percentage of time the CPU is busy
Scheduling Metrics: System
• Other
• fairness: no process should suffer starvation
• enforcing priorities: higher priority processes should not wait
First Come First Served (FCFS)
• Non-preemptive
• Implementation
• a queue of processes
• new processes enter the ready queue at the end
• when a process terminates the CPU is given to the process at the beginning of the queue
• (in practice) when a process blocks it goes to the end of the queue
Example
P1! P2! P3!
24! 27! 30!0!
Process Burst Time
P1 24
P2 3
P3 3
Suppose that the processes arrive in the order: P1 , P2 , P3. The Gantt Chart for the schedule is:
Two Metrics
• Average wait time to finish
• (d1 + (d1+d2) + (d1+d2+d3) + ... + ∑ dj )/n
• Average wait time to start
• (d1 + (d1+d2) + ... + ∑ dj )/n
P1 P2 P3 Pi Pn
d1 d2 d3 di dn
time
n
1
1
n-1
Analysis
• Total time to finish: 24+3+3 = 30
• Average time to finish (24+(24+3)+(24+3+3))/3 = 27
• Average time to start ((24)+(24+3))/3 = 17
P1! P2! P3!
24! 27! 30!0!
Analysis
• Total time to finish: 3+3+24 = 30
• Average time to finish (3+(3+3)+(3+3+24))/3 = 13
• Average time to start (3+(3+3))/3 = 3
P1!P3!P2!
6!3! 30!0!
Performance of FCFS• Pro:
• very simple code, data-structures and hence low overhead
• Con:
• can result int large average waiting times
• convoy effect
• General disadvantage due to lack of preemption
• consider a set of tasks with one long-running CPU intensive process and several I/O intensive tasks
Shortest Job First (SJF)
• Associate with each process the length of its next CPU burst
• Use these lengths to schedule the process with the shortest time
• FCFS is used to break ties
Example
Process Burst Time
P1 6
P2 8
P3 7
P4 3
P4! P3!P1!
3! 16!0! 9!
P2!
24!
Analysis
• Total time to finish: 6+8+7+3 = 24
• Average time to finish (3+(3+6)+(3+6+7)+(3+6+7+8))/4 = 13
• Average time to start (3+(3+6)+(3+6+7))/4 = 7
P4! P3!P1!
3! 16!0! 9!
P2!
24!
Performance of SJF
• Pro:
• if times are accurate, SJF gives a minimum average waiting time
• SJF is optimal
• Con:
• it is difficult to estimate CPU burst times
• Interesting as a lower limit on waiting time
Predicting the CPU Burst
• For long-term scheduling
• the user can be “encouraged” to give an estimate
• part of the job submission requirements
Predicting the CPU Burst
• For short-term scheduling
• we can attempt to predict its value by assuming some locality in CPU bursts and use exponential averaging
• τn+1 = α * Tn + (1 - α) * τn
• τn is the estimated value for the nth CPU burst
• Tn is the actual most recent burst value
• 0 ≤ α ≤ 1, α = 0.5 is commonly used
• the estimate lags the (potentially) sharper transitions of the CPU bursts
Predicting the CPU Burst
Modifications to SJF• Preemptive SJF
• if the shortest estimated CPU burst among all the processes in the ready queue (say this belongs to Pj) is less than the remaining time for the one that is running,
• preempt the currently running job;
• use its remaining time as its next CPU burst and add it to the ready queue
• start process Pj
• Shortest-Remaining-Time-First (SRTF)
• Policy prioritizes jobs with short CPU bursts
Example
• Average waiting time = ((10-1)+(1-1)+(17-2)+5-3))/4 = 6.5
P1! P1!P2!
1! 17!0! 10!
P3!
26!5!
P4!
Process Arrival Time
Burst Time
P1 0 8
P2 1 4
P3 2 9
P4 3 5
Priorities: A More General Notion
• Assign a numerical priority to each process
• convention: a smaller number means higher priority
• Examples of priority use
• value of the next CPU burst
• importance of interactive responsive
Priorities: A More General Notion
• Priorities can be based upon external considerations
• the importance of the user group running the process
• the amount of economic investment that the group might have in the system
• or upon internal considerations
• memory and other needs of the job
• ratio of CPU to I/O burst times
• number of open files etc.
Priorities: A More General Notion
• Priority-based scheduling
• assign the CPU to the process with highest priority
• may be used with or without preemption
• SJF is priority scheduling where priority is the inverse of predicted next CPU burst time
Example
• Average waiting time = 8.2
P2! P3!P5!
1! 18!0! 16!
P4!
19!6!
P1!
Process Burst Time
Priority
P1 10 3
P2 1 1
P3 2 4
P4 1 5
P5 5 2
Disadvantages of Priority Schemes
• A process can continuously be overtaken by higher priority processes arriving later
• can lead to starvation
• leads to better overall performance perhaps
• but not from the point of view of the process in question
• happens in real OSes unless special measures are taken
Disadvantages of Priority Schemes
• Common solution
• a process' priority goes up with its age
• FCFS is used to break ties between processes with equal priorities
• a process will not wait forever
• given enough time in the ready queue, its priority will eventually be the highest
• What should happen if low-priority process holds resources require by the high-priority process? (priority inversion)
Example of Priority Aging: Unix
• Priority goes up with lack of CPU usage
• process accumulates CPU usage
• Recalculates priority every time unit (~ 1 second)
• priority = CPUusage + basepriority
• CPUusage = (CPUusage) / 2
• A smaller number implies a higher priority
• basepriority is defined by the user (predefined range)
Round Robin (RR)
• Algorithm
• choose a fixed time unit, called a quantum
• allocate CPU time in quanta
• preempt the process when it has used its quantum
• typically, FCFS is used as a sequencing policy
Round Robin (RR)
• Each new process is added at the end of the ready queue
• When a process blocks or is preempted, it goes to the end of the ready queue
• A strictly preemptive policy
• very common choice for scheduling interactive systems
Round Robin (RR)
Example
P1! P2! P3! P1! P1! P1! P1! P1!
0! 4! 7! 10! 14! 18! 22! 26! 30!
Process Burst Time
P1 24
P2 3
P3 3
Typically, higher average turnaround than SJF, but better response
Quantum Size• Quantum size q is critical
• Affects waiting and turnaround times
• if q is the quantum size and there are n processes in the ready queue,
• The maximum wait is (n-1) * q units of time
• As q increases, we approach FCFS scheduling
• As q decreases the rate of context switches increases
• the average wait time goes down, and the system approaches one with 1/n the speed of the original system
Quantum Size
Quantum Size
80% of CPU bursts should be shorter than q
Multilevel Queue Scheduling
• Processes are partitioned into groups based on static criteria
• background (batch)
• foreground (interactive)
• All the processes in a fixed group of the partition share the same scheduling strategy and a distinct family of queues
• different scheduling algorithm can be used across different groups
• background: FCFS
• foreground: Round Robin
Multilevel Queue Scheduling
• Need to schedule the CPU between the groups as well
• fixed-priority: e.g., serve all from foreground, then from background
• possibility of starvation
• time slice: each group gets a certain fraction of the CPU
• e.g., 80% to foreground in RR, 20% to background in FCFS
Multilevel Queue Scheduling
Multilevel Feedback Queues
• Provide a mechanism for jobs to move between queues
• aging can be implemented this way
• Complete specification
• queues: number, scheduling algorithms (within and across queues)
• promotion and demotion policies
• which queue should a process enter when it needs service?
Example
• 3 queues: Q0 (FCFS, 8ms), Q1 (FCFS, 16ms), Q2 (FCFS)
• scheduling:
• process starts in Q0, gets 8ms of the CPU (FCFS)
• if not done, it moves to Q1 where it gets a further 16 ms of the CPU (FCFS)
• if still not done, it moves to Q2 and stays there till completed
Example
Choosing a Scheduling Approach
• Characterize system components as random processes
• random variables for CPU, I/O bursts
• typically have exponential distributions
• random variable for process arrival time
• we are interested in the average values of metrics
• e.g., average throughput, utilization
Modeling
• System modeled as a network of servers
• Queueing theory:
• for a queue, Little’s formula is n = λ W
• where λ is the arrival rate; W is the average waiting time; and n is the average queue length
• Has limitations on the degree to which realistic systems can be modeled and analyzed
Simulations
• When analytical solutions are infeasible, the models are simulated
• distribution driven as in the previous case,
• or actual (possibly synthetic) workloads
• Leads to more realistic predictions than the above forms of analysis
• Computationally more expensive than either of the above approaches
Actual Measurements
• Instrument a real system and capture the extra information
• Can make very precise determination
• Very cumbersome and time-consuming to achieve
• a problem is that observation changes the system
