MM Process Management

Karrie KarahaliosSpring 2007

(based off slides created by Brian Bailey)

Announcements

CPU Scheduling

• Maintain many programs in memory

• Determine how to best schedule them

• Requires information about the processes and an understanding of the system goals

• Switch among processes rapidly so each user perceives direct ownership of system

Process State

• New – being created

• Terminated – finishing execution

• Running – executing on the CPU

• Waiting – blocking on an event

• Ready – waiting for the CPU

Process Control Block

• Represents a process in the OS– process state– program counter– CPU registers– scheduling information– virtual memory information (e.g., page tables)– I/O status information

Resource Queues

Sem Queue

head

tail

Ready Queue

head

tail

PCB6

State

Registers

…

Disk Queue

head

tail

PCB5

State

Registers

…

PCB4

State

Registers

…

PCB3

State

Registers

…

PCB2

State

Registers

…

PCB1

State

Registers

…

CPU Scheduler

• When CPU becomes idle, select next process from the ready queue

• CPU may become idle when:– allotted time slice expires– interrupt occurs (timer, I/O, user input)– application makes an I/O request– application terminates

Evaluation Criteria

• Throughput: processes completed per unit time

• Turnaround: from submission to termination

• Wait: time waiting in the ready queue

• Response: from user request to system response

• Utilization: how busy the CPU is over time

• Want predictable system behavior

Scheduling Algorithms

• FCFS

• Shortest Job First

• Priority scheduling

• Round-robin

• Multi-level queue

• Rate monotonic

• Earliest deadline first

FCFS

• Process requests in order of arrival– non-preemptive algorithm– fairness (human vs. system perspective)– convoy effect

Shortest-Job-First

• Schedule process with smallest burst– difficult to know next CPU burst– use weighted average of past performance

• Gives min wait time for processes– if preemptive, may lead to starvation

nnn t )1(1

Priority

• Associate priority with each process

• Schedule process with highest priority– equal priority resolved via FCFS– if priority is a function of time, gives SJF

• Combat CPU starvation using aging– increase priority of long-waiting processes

Round Robin

• Schedule process after a time quantum– circulate around queue, give each process

single time quantum to execute– results in lots of context switches

• Selection of time quantum– short quantum results in longer waits– infinite quantum results in FCFS– should be large relative to context switch time

Take Home Exercise

• Assume workload shown on the right

• Compare FCFS, SJF, and RR (quantum=10)– compare along each

evaluation criteria

Process Arrival CPU Burst

P1 1 10

P2 2 29

P3 3 3

P4 4 7

P5 5 12

Multimedia Processes

• Exhibit continuous, periodic behavior

• Execute along with other processes

• Have perceptual constraints (QoS)

Scheduling

• Determine sharing of a resource such that all admitted tasks meet their deadlines

• Balance need to meet deadlines with maximizing throughput and utilization

Admissibility Test

• If test succeeds, the task can be added, and all tasks will still meet deadlines– called a feasible schedule

• Process utilization (U)<1 if deadlines considered

n

i

ii TCU1

/

Rate Monotonic Scheduling

• Static, fixed priority-based, periodic tasks– Ci is execution time of task i

– Ti is request period of task i

• Determines if a set of periodic tasks can be scheduled or not (feasible schedule)– no other fixed priority scheme can schedule a

set of tasks which can’t be scheduled by RM

Liu, C.L. and J.W. Layland. Journal of the ACM, 20(1): 46-61, 1973

Model Assumptions

• All requests to process tasks are periodic

• Each task must finish before next request

• Each request is independent

• Execution time for each task is constant

• Non-periodic tasks are special

• Tasks are preemptive based on priority– static: priorities assigned once at initialization– dynamic: priorities change during execution

Terminology

• Deadline– time of the next request for same task

• Feasible– no deadlines are missed

• Response time– time between request and completion of a task

• Critical instant– when a request for a task will have the largest response time– occurs when a task is requested simultaneously with requests

for all higher priority tasks• Critical time zone

– time between a critical instant and the end of the response to the corresponding request of the lowest priority task

Example 1

T1 = 2

C1 = 1 C2 = 1

T2 = 5

T1 is higher priority

Utilization (U) =

Example 2

T1 = 2

C1 = 1 C2 = 2

T2 = 5

T1 is higher priority

Utilization (U) =

Feasibility Test

• If requests for all tasks at their critical instants can be fulfilled, then feasible

• Determine least upper bound of U– maximum utilization such that all deadlines

are still satisfied, in the general case

• Assigning highest priority to task with the smallest period yields best utilization

Problem for Two Tasks

• Tasks (C1, T1) and (C2, T2), with T1 < T2

• Within request period of T2, there can be at most ceil(T2 / T1) requests from T1

• Adjust C2 to fully utilize CPU

Case 1

• Requests for task1 completed within T2

– no overlap with T2

• In this case, largest value of C2 can be

C2 = T2 – C1*ceil(T2/T1)

• U = 1 + C1[ (1/T1) – (1/T2) *ceil(T2 / T1)]

Case 2

• Execution of ceil(T2/T1)th request from task1 overlaps second request from task2

– e.g. (C1=1,T1=1.5); (C2=1, T2=5)

• In this case, largest value of C2 can be

C2 = T1*fl(T2/T1) – C1*fl(T2/T1)

• U = (T1/T2)*fl(T2/T1) + C1[ (1/T1) – (1/T2) *fl(T2 / T1)]

Minimum Occurs at Boundary

• Minimum of U occurs at the boundary between these two cases

• Set equations equal and solve for C1

C1 = T2 – T1*fl(T2/T1)

• U = 2(21/2 – 1) =~ 0.83– highest utilization such that all deadlines met

General Case

• For two tasks– U < 2 (2½ - 1), ~= 0.83

• For three tasks– U < 3 (21/3 - 1) ~= 0.78

• For n tasks– U < n (21/n – 1) -> ln 2 ~= 0.69

• If configuration of tasks pass the test, no deadlines will be missed– must limit utilization to 69% for large N !!!

Example

• Is this configuration of tasks admissible?

C1 = 1.5 C2 = 2.1

T1 = 2.5 T2 = 6

Example

• Is this configuration of tasks admissible?

C1 = 1 C2 = 2

T1 = 2 T2 = 4

Deadline Driven Scheduling (EDF)

• Assign priorities to tasks based on the deadlines of their current requests

• Execute task with highest priority and yet unfulfilled request

• Optimal scheduling algorithm– If tasks can be scheduled by any algorithm,

then they can also be scheduled by EDF

Compare Rate Monotonic vs. EDF

– C1 = 1.5 C2 = 2.1

– T1 = 2.5 T2 = 6

• Draw timing models for each– show failure with RM, but success with EDF