8: Scheduling:

The Multi-Level Feedback Queue

Operating System: Three Easy Pieces

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 1

No more Oracle

 Finally let’s relax our final assumption - that the scheduler knows the
length of each job.

 How can we build an approach that behaves like SJF/STCF without

such a priori knowledge

 Simultaneously how can we incorporate some of the ideas we have

seen with the RR scheduler so that response time is also quite good?

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 2

Multi-Level Feedback Queue (MLFQ)

 A Scheduler that learns from the past to predict the future.

 Objective:
 Minimize response time for interactive jobs

 Optimize turnaround time without a priori knowledge of job length.

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 3

MLFQ: Basic Rules

 MLFQ has a number of distinct queues.

 Any time a job that is ready to run is in one of the queues

 Each queues is assigned a different priority level.

 Uses priority to decide which job should run at given time.

 A job on a higher queue is chosen to run.

 Use round-robin scheduling among jobs in the same queue

Rule 1: If Priority(A) > Priority(B), A runs (B doesn’t).

Rule 2: If Priority(A) = Priority(B), A & B run in RR.

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 4

MLFQ: Basic Rules (Cont.)

 How to know how to set priority?

 MLFQ varies the priority of a job based on its observed behavior.

 Example:
 A job repeatedly relinquishes the CPU while waiting IOs → Keep its priority
high

 A job uses the CPU intensively for long periods of time → Reduce its
priority.

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 5

MLFQ Example

[High Priority] Q8 A B

Q7

Q6

Q5

Q4 C

Q3

Q2

[Low Priority] Q1 D

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 6

MLFQ: How to Change Priority

 MLFQ priority adjustment algorithm:

 Rule 3: When a job enters the system, it is placed at the highest priority

 Rule 4a: If a job uses up an entire time slice while running, its priority is
reduced (i.e., it moves down on queue).

 Rule 4b: If a job gives up the CPU before the time slice is up, it stays at
the same priority level

In this manner, MLFQ approximates SJF

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 7

Example 1: A Single Long-Running Job

 A three-queue scheduler with time slice 10ms

Q2

Q1

Q0

0 50 100 150 200

Long-running Job Over Time (msec)

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 8

Example 2: Along Came a Short Job

 Preemption:
 Job A: A long-running CPU-intensive job

 Job B: A short-running interactive job (20ms runtime)

 A has been running for some time, and then B arrives at time T=100.

Q2 A:

Q1 B:

Q0

0 50 100 150 200

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 9

Example 3: What About I/O?

 Assumption:
 Job A: A long-running CPU-intensive job

 Job B: An interactive job that need the CPU only for 1ms before
performing an I/O

Q2 A:

Q1 B:

Q0

0 50 100 150 200

A Mixed I/O-intensive and CPU-intensive Workload (msec)

The MLFQ approach keeps an interactive job at the highest priority

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 10

Problems with the Basic MLFQ

 Starvation
 If there are “too many” interactive jobs in the system.

 Lon-running jobs will never receive any CPU time.

 Game the scheduler

 After running 99% of a time slice, issue an I/O operation.

 The job gain a higher percentage of CPU time.

 A program may change its behavior over time.

 CPU bound process → I/O bound process

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 11

The Priority Boost

 Rule 5: After some time period S, move all the jobs (haven’t been
scheduled) in the system to the topmost queue.
 Example:
 A long-running job(A) with two short-running interactive job(B, C)

A: B: C:

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 12

Better Accounting

 How to prevent gaming of our scheduler?

 Solution:
 Rule 4 (Rewrite Rules 4a and 4b): Once a job uses up its time allotment at
a given level (regardless of how many times it has given up the CPU), its
priority is reduced(i.e., it moves down on queue).

Q2 Q2

Q1 Q1

Q0 Q0

0 50 100 150 200 0 50 100 150 200

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 13

Tuning MLFQ And Other Issues

Lower Priority, Longer Quanta

 The high-priority queues → Short time slices

 E.g., 10 or fewer milliseconds

 The Low-priority queue → Longer time slices

 E.g., 100 milliseconds

Q2

Q1

Q0

0 50 100 150 200

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 14

The Solaris MLFQ implementation

 For the Time-Sharing scheduling class (TS)

 60 Queues

 Slowly increasing time-slice length

 The highest priority: 20msec

 The lowest priority: A few hundred milliseconds

 Priorities boosted around every 1 second or so.

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 15

MLFQ: Summary

 The refined set of MLFQ rules:

 Rule 1: If Priority(A) > Priority(B), A runs (B doesn’t).

 Rule 2: If Priority(A) = Priority(B), A & B run in RR.

 Rule 3: When a job enters the system, it is placed at the highest priority.

 Rule 4: Once a job uses up its time allotment at a given level (regardless
of how many times it has given up the CPU), its priority is reduced(i.e., it
moves down on queue).

 Rule 5: After some time period S, move all the jobs in the system to the
topmost queue.

Operating Systems by Dr. Praveen Kumar @ CSED, VNIT Nagpur 16