Real-Time Task Model

- Chapter 3 -

“Monolithic approach”
Good enough?
• No!
– For a complex system, it is really hard to design a
single superloop
– Really hard to validate cross-related temporal
requirements
• So, we need a more structured way to look at the
problem
– Reference model: abstracted view (look at only core) of
• Workload
• Resource
– Easy to form a structured way of design and validation
– Can be generalized for many variations of problems

1
 Why Reference Model?
LeastSquare
Kalmanfilter P
filter PID
control Abstraction
control
Specific
ARM
Problem 1 Model
processor
Problem
sensor actuator Specific
sensor actuator
actuator Problem 2
sensor

Customizing

The Objective of A Reference Model

• Directly handling wide variety of details (Kalman or
Sorting, C++ or Ada, Pentium III or ARM-7, Unix or
Windows)?
– have to deal with irrelevant details
– hard to be generalized
• A Reference Model captures only relevant characteristics
and categorize the problem space
– workload model
– resource model
– algorithms
• Like doctor’s reference handbook, a reference model help
you
– classify a given real-time problem
– know if it is a solved problem or open problem
– know if it is a easy problem or hard problem

2
 Overview
• Workload model (characterization of applications)
– Jobs and tasks
– Release time (periodic/aperiodic/sporadic)
– Deadline (absolute/relative, hard/soft, single stage/end to end)
– Execution time (deterministic/stochastic)
– Temporal distance and precedence constraints
• Resource model
– Processors and resources
– Utilization of resource
• Scheduling algorithms
– Priority/preemptability/blocking
– Scheduler and schedule

Workload Model

0 (system startup time)

• Task and job

• Task offset
• Release time of jobs (periodic, sporadic, aperiodic)
• Deadline of jobs (hard or soft)
• Execution time of jobs (deterministic or stochastic)

3
 Jobs and Tasks
• A job is a unit of computation, e.g.,
– handling the press of a keyboard
– or compute the control response in one instance of a
control loop

• A task is a sequence of the same type of jobs, say,

a control task or the keyboard handling task.

job1 Job 2

Release Time and Deadline

•Release time is the instant at which the job becomes ready to
execute
•Deadline is the time by which the job should complete.

job1 Job 2

deadline
Release
time

4
 Periodic, Aperiodic and
Jitter in Release times
• A task is periodic, if the release times are periodic, e.g.,
every second.

• But realistically, we cannot do it exactly every second in a

physical system. There could be some small errors such as
a window of “ ± 1 msec”. This is known as jitter

• If the release time is irregular. It is known as aperiodic. If

the aperiodic arrivals have a minimal distance that
separates two consecutive release times, it is also known as
sporadic task. (This term is not universally agreed.)

Absolute and Relative Deadlines

• The common form of deadlines are absolute deadlines
where deadlines are specified in, well, absolute times.
Train and airlines schedules are absolute but soft deadlines.
• Normally, relative deadlines are related to the release time.
For example, 10 msec after the release time.
• Sometimes, it is specified to be n units of time after an
event is detected. For example, upon the detection of a
frontal collision, the airbag must be deployed within k
msec. In this case, the occurrence of the event is the release
time.
• Laxity is a measure of how long is left to hit the deadline,
say, 10 sec left to launch the Shuttle.

5
 Hard and Soft Deadlines
• Deadlines are considered to be hard if failing to meet them
is considered as an application failure leading to
catastrophic consequences on the controlled environment.
• A deadline is considered to be soft, if it is specified
probabilistically. For example, a radar system’s ability to
detect and display a plane by some given time is usually
specified in the following style:
– When a plane is over the horizon, it must be detected and
displayed on operator’s screen no later than 1 second 99.95 % of
the times.
• A deadline is also considered to be soft if a late completion
gracefully degrades the performance of the system without
causing damage to the controlled environment

Deterministic and Stochastic

Execution Times
• Literally speaking, there is no such thing as “deterministic
execution times”. There will always be some variations.
• Deterministic execution time in practice means that we
know the worst-case execution time. In practice, it often
implies that the worst-case execution time is not too far
away from average execution time so that it is practical to
use it. Most numerical computation used in control fits this
model.
• Stochastic execution time in practice means that the use of
the worst-case execution time is no longer practical since it
is too far off from the average. Compressed motion video
is a good example.

6
 Periodic Task Model
• Periodic tasks are the “work horse” of real-time
systems and they play a key role in real-time
systems.

• A periodic task, Ti, is characterized by

– Phase (or offset) φi : the starting time of the task, i.e.,
the first release time (ri,1). The default is 0.
– Period, pi
– Release time, ri,j. The default is ri,j = ri,j-1 + pi
– Execution, Ci. The default is worst-case execution time
– Relative deadline Di. The default is end of period

Temporal Distance - 1
• Sequential temporal distance constraint: how far the
completion time of two consecutive jobs can be separated.
• Example Application: Advanced phase array radar.
Temporal distance and deadline specify the window of
time at which another look (SAMPLING) of a target is
needed. Temporal distance and deadline are computed
online according to the nature and speed of the target.
• Quiz: why not using a periodic model?
relative deadline

Job Job Job

Temporal
distance

7
 Temporal Distance - 2
• Periodic model is still in use. But it permits bundling,
which wastes antenna resource. If next sample completes
at the end of next period, the wide separation will result in
more error.
• Instance by instance temporal distance (from the current
job’s completion time to the completion time of next job)
gives more precise control on the use of the antenna.
bundling

Precedence Constraints
• Precedence relation says you can’t start until your
predecessor has done.

• Example: “Data producer job” has a precedence

constraint with “Data consumer”. For a sensory
data fusion, filtering of individual sensor data
should be completed before their fusion.

8
 Precedence Constraint Representation
(Task Graph)
(0,7] (2,9] (4,11] (6,13] (8,15]
…
(2,5] (5,8] (8,11] (11,14] (14,17]
…
(0,5] (4,8] (5,20] conditional block

branch join
(0,6]

(2,10]

2/3 1/2

Resource Model
(Processors and Resources)
• Active resources: processor, bus, disks and communication
channels where a job executes or transmits over it. The
size of the job and the speed of the active resources
determine the execution/transmission time. Usually, active
resources are scheduled, for example, scheduling the CPU
or LAN.

• Passive resources or just resources: shared variables,

semaphores, or library routine etc, which the jobs need to
make execution (transmission). We do not associate speed
with a passive resource.

9
 Utilization of Active Resource
by a task
• A periodic task’s utilization Ui of an active resource is the
ratio between its execution time and period: Ui = Ci/pi

• Given a set of periodic tasks on an active resource, e.g. the

CPU, the CPU’s utilization is equal to the sum of periodic
tasks’ utilization:
Ci
U =∑
i pi

• Given a set of aperiodic tasks, the average CPU utilization

is the ratio of average job execution time and average job
inter-arrival time.

Scheduling Model
“Schedule and Scheduler”
• Given a set of jobs, J= {J1,…,Jn}, a schedule σ(t) is an assignment of
jobs to the resource, so that each task is executed until completion.
• σ(t) is an integer step function and σ(t)=k, with k>0, means that job Jk
is executing at time t, while σ(t)=0 means that the CPU is idle.

Context switch
idle idle
J1 J2 J3
σ(t)

t1 t2 t3 t4 t

10
 Schedule and Scheduler
• Scheduler is the module that implements the scheduling
algorithm. Each processor is assigned one job at a time.
– Jobs become ready to execute after their release time and after
requested resources are allocated and precedence constraints are
satisfied
– Jobs are executed in the priority order, preemption carries out and
processor is de-allocated immediate after a job is done.
• A schedule is said to be feasible if all tasks can be
completed according to a set of specified constraints.
• A set of tasks is said to be schedulable if there exists at
least one algorithm that can produce a feasible schedule for
the task set.

Preemptive vs Non-preemptive
scheduling
• Preemption: A preemptive schedule is a schedule
in which the running task can be arbitrarily
suspended at any time, to assign the CPU to
another task according to a predefined scheduling
policy

• CPU is preemptable.
• Communication channel is non-preemptable.
• Passive resource, e.g., shared data, is not
preemptable – mutually exclusive.

11
Classification of Scheduling Algorithms

• Preemptive/ Non-preemptive. In preemptive scheduling,

the running task can be interrupted at any time to assign
the processor to another active task. On the other hand, in
non-preemptive scheduling, a task, once started, is
executed by the processor until completion.
• Off-line/On-line/Clairvoyant. The scheduling algorithm
runs off-line/on-line. In the off-line case, the generated
schedule is stored in a table and later executed by a
dispatcher. In on-line case, the schedule is determined
online by looking at jobs in the ready queue. Clairvoyant
(like GOD!) know even the future jobs and schedule jobs
accordingly.

12