Chapter 3: Real-Time Scheduling and Schedulability Analysis

Albert M. K. Cheng

Real-Time Scheduling Schedulers: compile-time, run-time Systems: uniprocessor, multiprocessor Characterization of a task:

c computation time (WCET)

s start (release or ready) time

d deadline (relative to start time)

p period or minimum separation

Optimal Scheduler One which may fail to meet the deadline of

task only if no other scheduler can Examples include uniprocessor earliest-

deadline-first (EDF) and least-laxity-first (LLF) schedulers for independent tasks with no synchronization constraints and no resource requirements

EDF vs FIFO (non-real-time scheduler)Example: task s d c

1 0 5 3

2 0 12 4

3 0 10 4

ready queue: head 2, 3, 1 tail

FIFO schedule: 2, 3, 1 (task 1 misses deadline)

EDF schedule: 1, 3, 2 (all tasks meet deadlines)

Rate-Monotonic (RM) Scheduler:Basic Assumptions Compile-time/static Preemptive tasks No precedence constraints Negligible context-switching time Periodic tasks

Schedulability Test for RM Liu and Layland’s sufficient condition:

Utilization U = sum c_I / p_I

Number of tasks n

A task set is schedulable if

U <= n(2^(1/n) – 1)

Necessary and sufficient condition requires

checking inequalities or plot of W vs t

Sporadic Tasks Can be modeled by periodic tasks with

periods equal to the minimum separation Deferred server (DS) is less wasteful Tasks with critical sections (non-

preemptive) must be handled differently

Scheduling Tasks with Constraints Deterministic Rendezvous Model: tasks

with rendezvous (as in Ada) Kernelized Monitor Model: tasks with

critical sections Dataflow Model: tasks with precedence

and data flow constraints

Multiprocessor Scheduling Scheduling game board representation:

shows a tasks’ statuses dynamically Non-optimality of uniprocessor-optimal

schedulers Need to consider special cases to improve

scheduling efficiency Need to avoid conflict-free task sets