Sharif University of Technology

Department of Computer Science and Engineering

Lec. 2:
Basic Concepts

Real-Time Computing

S. Safari
Spring 2024
 Introduction

§ A process is a computation that is executed by the CPU in

a sequential fashion.
• In this course, the term process is used as synonym of
task and thread.
§ Concurrent tasks means tasks that can overlap in time.
• When a single processor has to execute a set of concurrent tasks
the CPU has to be assigned to the various tasks according to a
predefined criterion, called a scheduling policy.
• The set of rules that, at any time, determines the order in which
tasks are executed is called a scheduling algorithm.
• The specific operation of allocating the CPU to a task selected by
the scheduling algorithm is referred as dispatching.

Sharif University of Technology - 2/48 -

 Introduction (2)

§ A task that could potentially execute on the CPU can be:

• In execution (if it has been selected by the scheduling algorithm).
• Waiting for the CPU (if another task is executing).

§ Active task: A task that can potentially execute on the

processor, independently on its actual availability.
§ Ready task: A task waiting for the processor.
§ Running task: The task in execution.

Sharif University of Technology - 3/48 -

 Introduction (3)

§ All ready tasks waiting for the processor are kept in a

queue, called ready queue.
• Operating systems that handle different types of tasks may have
more than one ready queue.
• The operation of suspending the running task and inserting it into
the ready queue is called preemption.

Sharif University of Technology - 4/48 -

 Preemption

§ In dynamic real-time systems, preemption is important for

three reasons [SZ92]:
• Tasks performing exception handling may need to preempt
existing tasks so that responses to exceptions may be issued in a
timely fashion.
• When tasks have different levels of criticality (expressing task
importance), preemption permits executing the most critical tasks,
as soon as they arrive.
• Preemptive scheduling typically allows higher efficiency, in the
sense that it allows executing real-time task sets with higher
processor utilization.
§ Preemption destroys program locality and introduces a runtime
overhead that inflates the execution time of tasks.

Sharif University of Technology - 5/48 -

 An Example of Schedule

§ An example of schedule obtained by executing three tasks: J1, J2, J3.

§ At times t1, t2, t3, and t4, the processor performs a context switch.
§ Each interval [ti, ti+1) in which σ(t) is constant is called time slice.
Interval [x, y) identifies all values of t such that x ≤ t < y.

Sharif University of Technology - 6/48 -

 An Example of A Preemptive Schedule

§ 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.

Sharif University of Technology - 7/48 -

 Feasible Schedule and Schedulable Tasks

§ In preemptive schedules, tasks may be executed in

disjointed interval of times.
§ 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.

Sharif University of Technology - 8/48 -

 Types of Task Constraints

§ Typical constraints that can be specified on real-time tasks

are of three classes:
• Timing constraints
• Precedence relations
• Mutual exclusion constraints on shared resources

Sharif University of Technology - 9/48 -

 Timing Constraints

§ A typical timing constraint on a task is the deadline.

• Represents the time before which a process should complete its
execution without causing any damage to the system.
• Relative deadline: If a deadline is specified with respect to the task
arrival time.
• Absolute deadline: if it is specified with respect to time zero.

Sharif University of Technology - 10/48 -

 Categories of Real-Time Tasks

§ Depending on the consequences of a missed deadline,

real-time tasks are usually distinguished in three
categories:
• Hard: A real-time task is said to be hard if missing its deadline may
cause catastrophic consequences on the system under control.
• Firm: A real-time task is said to be firm if missing its deadline does
not cause any damage to the system, but the output has no value.
• Soft: A real-time task is said to be soft if missing its deadline has
still some utility for the system, although causing a performance
degradation.

Sharif University of Technology - 11/48 -

 Characterizations of A Real-Time Task

§ Arrival time ai: The time at which a task becomes ready for
execution; it is also referred as request time or release time
and indicated by ri.
§ Computation time Ci: The time necessary to the processor
for executing the task without interruption.
§ Absolute Deadline di: The time before which a task should
be completed to avoid damage to the system.
§ Relative Deadline Di: The difference between the absolute
deadline and the request time: Di = di − ri.

Sharif University of Technology - 12/48 -

 Characterizations of A Real-Time Task (2)

§ Start time si: The time at which a task starts its execution.
§ Finishing time fi: The time at which a task finishes its
execution.
§ Response time Ri: The difference between the finishing time
and the request time: Ri =fi−ri.
§ Criticality is a parameter related to the consequences of
missing the deadline (typically, it can be hard, firm, or soft).
§ Value vi represents the relative importance of the task with
respect to the other tasks in the system.

Sharif University of Technology - 13/48 -

 Characterizations of A Real-Time Task (3)

§ Lateness Li: Li=fi−di represents the delay of a task

completion with respect to its deadline.
• If a task completes before the deadline, its lateness is negative.

§ Tardiness or Exceeding time Ei: Ei=max(0, Li) is the time a

task stays active after its deadline.
§ Laxity or Slack time Xi: Xi=di−ai−Ci is the maximum time a
task can be delayed on its activation to complete within its
deadline.

Sharif University of Technology - 14/48 -

 Characterizations of A Real-Time Task (4)

§ Some of the parameters defined in the previous slides are illustrated in

the following figure.

Sharif University of Technology - 15/48 -

 The Regularity of Activation

§ Tasks can be defined as periodic or aperiodic.

§ Periodic tasks consist of an infinite sequence of identical activities,
called instances or jobs, that are regularly activated at a constant rate.
§ Aperiodic tasks also consist of an infinite sequence of identical jobs (or
instances);
• Their activations are not regularly interleaved.
§ An aperiodic task where consecutive jobs are separated by a minimum
inter-arrival time is called a sporadic task.

Sharif University of Technology - 16/48 -

 The Regularity of Activation (2)

§ A periodic task τi; an aperiodic job Ji; the generic kth job of a periodic task τi,k.
§ The activation time of the first periodic instance (τi,1) is called phase.
• The activation time of the kth instance is given by φi+(k−1)Ti.
• Note: Ti is the activation period of the task. A periodic process can be completely
characterized by its phase φi, its computation time Ci, its period Ti, and its relative
deadline Di.

Sharif University of Technology - 17/48 -

 Precedence Constraints

§ Precedence relations are usually described through a directed

acyclic graph G.
• Tasks are represented by nodes and precedence relations by arrows
§ The notation Ja≺Jb specifies that task Ja is a predecessor of task
Jb, meaning that G contains a directed path from Ja to Jb.
§ The notation Ja→Jb specifies that task Ja is an immediate
predecessor of Jb, meaning that G contains an arc directed from
node Ja to node Jb.
§ Tasks with no predecessors:
beginning tasks;
§ Tasks with no successors:
ending tasks

Sharif University of Technology - 18/48 -

 Resource Constraints

§ From a process point of view, a resource is any software structure

that can be used by the process to advance its execution.
§ A resource dedicated to a particular process is said to be private,
whereas a resource that can be used by more tasks is called a
shared resource.
§ Some resources:
• A data structure, a set of variables, a main memory area, a file, a
piece of program, or a set of registers of a peripheral device.

Sharif University of Technology - 19/48 -

 A Mutually Exclusive Resource

§ Many shared resources do not allow simultaneous accesses by

competing tasks, but require their mutual exclusion.
• This means that a task cannot access a resource R if another task is
inside R manipulating its data structures.
• In this case, R is called a mutually exclusive resource.
• A piece of code executed under mutual exclusion constraints is called
a critical section.

Sharif University of Technology - 20/48 -

 An Example of Mutual Exclusion

§ Two tasks cooperate to track a moving object: task τW gets the

object coordinates from a sensor and writes them into a shared
buffer R, containing two variables (x, y); task τD reads the variables
from the buffer and plots a point on the screen to display the
object trajectory.

Sharif University of Technology - 21/48 -

 Solution: Semaphores

Sharif University of Technology - 22/48 -

 Waiting State

§ When a running task executes a wait primitive on a locked

semaphore, it enters a waiting state, until another task executes a
signal primitive that unlocks the semaphore.
• Note that when a task leaves the waiting state, it does not go in the
running state, but in the ready state, so that the CPU can be
assigned to the highest-priority task by the scheduling algorithm.

Sharif University of Technology - 23/48 -

 Definition of Scheduling Problems

§ A set of n tasks Γ = {τ1, τ2, . . . , τn},

§ A set of m processors P = {P1, P2, . . . , Pm}
§ A set of s types of resources R = {R1, R2, . . . , Rs}
§ Precedence relations among tasks can be specified through a
directed acyclic graph, and timing constraints can be associated
with each task.
§ Scheduling means assigning processors from P and resources
from R to tasks from Γ in order to complete all tasks under the
specified constraints.
§ This problem, in its general form, has been shown to be NP-
complete [GJ79] and hence computationally intractable.

Sharif University of Technology - 24/48 -

 Time Complexity of Online Scheduling Problems

§ The complexity of scheduling algorithms is of high relevance in

dynamic real-time systems.
• Scheduling decisions must be taken online during task execution.
§ A polynomial algorithm is one whose time complexity grows as a
polynomial function p of the input length n of an instance.
• The complexity of such algorithms is denoted by O(p(n)).
• Each algorithm whose complexity function cannot be bounded
in that way is called an exponential time algorithm.
• The online scheduling problems must be solved in polynomial
time.

Sharif University of Technology - 25/48 -

 Classification of Scheduling Algorithms

§ Preemptive vs. Non-preemptive

• In preemptive algorithms, the running task can be interrupted at any
time to assign the processor to another active task, according to a
predefined scheduling policy.
• In non-preemptive algorithms, a task, once started, is executed by
the processor until completion. In this case, all scheduling decisions
are taken as the task terminates its execution.
§ Static vs. Dynamic
• Static algorithms are those in which scheduling decisions are based
on fixed parameters, assigned to tasks before their activation.
• Dynamic algorithms are those in which scheduling decisions are
based on dynamic parameters that may change during system
evolution.

Sharif University of Technology - 26/48 -

 Classification of Scheduling Algorithms (2)

§ Off-line vs. Online

• A scheduling algorithm is used off line if it is executed on the entire task
set before tasks activation. The schedule generated in this way is stored
in a table and later executed by a dispatcher.
• A scheduling algorithm is used online if scheduling decisions are taken
at runtime every time a new task enters the system or when a running
task terminates.

§ Optimal vs. Heuristic

• An algorithm is optimal if it minimizes some given cost function defined
over the task set. When no cost function is defined and the only concern
is to achieve a feasible schedule, then an algorithm is optimal if it is able
to find a feasible schedule, if one exists.
• An algorithm is heuristic if it is guided by a heuristic function in taking its
scheduling decisions. A heuristic algorithm tends toward the optimal
schedule, but does not guarantee finding it.

Sharif University of Technology - 27/48 -

 Guarantee-Based Algorithms

§ In hard real-time applications that require highly predictable

behavior, the feasibility of the schedule should be guaranteed in
advance.
• Before task execution
§ Checking the feasibility of the schedule before tasks’ execution
§ Assuming a worst-case scenario

Sharif University of Technology - 28/48 -

 Guarantee-Based Algorithms (2)

§ In static real-time systems, where the task set is fixed and

known a priori, all task activations can be precalculated offline,
and the entire schedule can be stored in a table that contains all
guaranteed tasks arranged in the proper order.
• At runtime, a dispatcher simply removes the next task from the table
and puts it in the running state.
§ The main advantage of the static approach is that the runtime
overhead does not depend on the complexity of the scheduling
algorithm. This allows very sophisticated algorithms to be used to
solve complex problems or find optimal scheduling sequences.
§ On the other hand, however, the resulting system is quite inflexible
to environmental changes; thus, predictability strongly relies on the
observance of the hypotheses made on the environment.

Sharif University of Technology - 29/48 -

 Guarantee-Based Algorithms (3)

§ Dynamic real-time systems (typically consisting of firm tasks)

• Tasks can be created at runtime
• The guarantee must be done online every time a new task is
created.

Sharif University of Technology - 30/48 -

 Domino Effect

§ The benefit of having a guarantee mechanism is that potential

overload situations can be detected in advance to avoid negative
effects on the system.
§ One of the most dangerous phenomena caused by a transient
overload is called domino effect.
• It refers to the situation in which the arrival of a new task causes all
previously guaranteed tasks to miss their deadlines.

Sharif University of Technology - 31/48 -

 Best-Effort Algorithms

§ In certain real-time applications, computational activities have soft

timing constraints.
• Missing soft deadlines do not cause catastrophic consequences, but
only a performance degradation.
§ For example: Typical multimedia applications
• The objective: Handling different types of information (such as text,
graphics, images, and sound) in order to achieve a certain quality of
service for the users
• Missing a deadline may only affect the performance of the system.
§ To efficiently support soft real-time applications, a best-effort approach
may be adopted for scheduling.
• There is no guarantee of finding a feasible schedule.
• Best-effort algorithms perform much better than guarantee-based
schemes in the average case.
Sharif University of Technology - 32/48 -
 Metrics for Performance Evaluation

§ The performance of scheduling algorithms is typically evaluated

through a cost function defined over the task set.
• For example, classical scheduling algorithms try to:
• Minimize the average response time, the total completion time, the
weighted sum of completion times, or the maximum lateness.

Sharif University of Technology - 33/48 -

 Example

§ The schedule minimizes the maximum lateness, but all tasks miss
their deadline. The schedule shown in Figure b has a greater
maximum lateness, but four tasks out of five complete before their
deadline.

Sharif University of Technology - 34/48 -

 Utility Functions

§ Utility function: The value associated with the task as a function of

its completion time.
• The benefit of executing a task may depend not only on the task
importance but also on the time at which it is completed

Sharif University of Technology - 35/48 -

 Scheduling Anomalies

§ Real-time computing is not equivalent to fast computing.

§ Richard’s anomalies were described by Graham in 1976 and refer to task
sets with precedence relations executed in a multiprocessor environment.

Theorem (Graham, 1976). If a task set is optimally scheduled on a

multiprocessor with some priority assignment, a fixed number of
processors, fixed execution times, and precedence constraints, then
increasing the number of processors, reducing execution times, or
weakening the precedence constraints can increase the schedule
length.

Sharif University of Technology - 36/48 -

 An Example of Graham’s Theorem

§ Let us consider a task set consisting of nine jobs J = {J1, J2, ... , J9},
sorted by decreasing priorities, so that Ji priority is greater than Jj
priority if and only if i < j.

Sharif University of Technology - 37/48 -

 An Example of Graham’s Theorem (2)

§ If this task set is executed on a platform with three processing

cores, where the highest priority task is assigned to the first
available core, the resulting schedule σ∗ is illustrated in following
figure, where the global completion time is tc = 12 units of time.

Sharif University of Technology - 38/48 -

 An Example of Graham’s Theorem: Number of
Processing cores increased
§ Considering four processing cores
§ Obtaining the schedule which is characterized by a global
completion time of tc = 15 units of time.

Sharif University of Technology - 39/48 -

 An Example of Graham’s Theorem: Computation
Times Reduced
§ One could think that the global completion time of the task set J
could be improved by reducing tasks’ computation times of each
task. However, reducing the computation time of each task by one
unit of time, the schedule length will increase with respect to the
optimal schedule σ∗, and the global completion time will be tc = 13.

Sharif University of Technology - 40/48 -

 An Example of Graham’s Theorem: Precedence
Constraints Weakened
§ Removing the precedence relations between job J4 and jobs J7 and J8
• Obtaining the schedule which is characterized by a global completion
time of tc = 16 units of time.

Sharif University of Technology - 41/48 -

 An Example of Graham’s Theorem: Anomalies
under Resource Constraints
§ Five jobs are statically allocated on two processors:
• Jobs J1 and J2 on processor P1, and jobs J3, J4 and J5 on processor P2 (jobs
are indexed by decreasing priority).
• Jobs J2 and J4 share the same resource in exclusive mode; hence their
execution cannot overlap in time.

§ The total completion time is tc = 17.

Sharif University of Technology - 42/48 -

 An Example of Graham’s Theorem: Anomalies
under Resource Constraints (2)
§ Reducing the computation time of job J1 on the first core, then J2 can
begin earlier and take the resource before J4. As a consequence, job J4
must now block over the shared resource and possibly miss its deadline.
§ The blocking time experienced by J4 causes a delay in the execution of J5
(which may also miss its deadline), increasing the total completion time of
the task set from 17 to 22.

Sharif University of Technology - 43/48 -

 An Example of Graham’s Theorem: Anomalies
under Resource Constraints (Increasing Speed)
§ A real-time application that is feasible on a given processor can become
infeasible when running on a faster processor.

Sharif University of Technology - 44/48 -

 An Example of Graham’s Theorem: Anomalies
under Non-Preemptive Scheduling
§ Tasks are assigned a fixed priority proportional to their relative deadline,
thus τ1 is the task with the highest priority and τ3 is the task with the
lowest priority.
§ If task set is executed with double speed S2=2S1, τ1 misses its deadline.
• This happens because, when τ1 arrives, τ3 already started its
execution and cannot be preempted (due to the non-preemptive
mode).

Sharif University of Technology - 45/48 -

 An Example of Graham’s Theorem: Anomalies
Using A Delay Primitive
§ Another timing anomaly can occur when tasks using shared resources
explicitly suspend themselves through a delay(T) system call, which
suspends the execution of the calling task for T units of time.

Sharif University of Technology - 46/48 -

 An Example of Graham’s Theorem: Anomalies
Using A Delay Primitive (2)
§ Another case in which the suspension of a task can also cause a longer
delay in a different task, even without sharing any resource.

Sharif University of Technology - 47/48 -

 Summary

§ Definition of Process, Scheduling, Ready Queue, and etc.

§ Preemption
§ Types of Tasks Constraints
§ Characterizations of Real-Time Tasks
§ Precedence and Resource Constraints

Sharif University of Technology - 48/48 -