20.01.05 Lecture 2, CS5270 1

The Real Time Computing Environment I

CS 5270 Lecture 2

20.01.05 Lecture 2, CS5270 2

A Conceptual Framework• The outside view:

– Closed system : verification

– Open system : Synthesis, schdulability analysis, …

• The inside view– Architecture– Interrupts– Task scheduling…..– Embedded software

• Modeling, analysis, verification are important at all layers!

20.01.05 Lecture 2, CS5270 3

The Closed System View

Computing system

Plant

SenseActuate

Extract a model (Timed Automaton) ; Verify if this closed system meets a given specification.

20.01.05 Lecture 2, CS5270 4

The Open System View

Plant

SenseActuate

For this open system , synthesize a controller (real time computing system) such that the closed system meets a given specification.

20.01.05 Lecture 2, CS5270 5

The Open System View

Actuate

For this open system , synthesize a controller (real time computing system) such that the closed system meets a given specification.

Plant

Sense

Computing system

20.01.05 Lecture 2, CS5270 6

The Outside View

• Timed Automata :– outside view

closed system Verification

• Model extraction• Specification of properties• Verification methods/tools.

20.01.05 Lecture 2, CS5270 7

The Inside View

Plant

Sense

Computing system

Actuate

What is inside the black box?

20.01.05 Lecture 2, CS5270 8

Sense

Computing system

Actuate

Sense

Actuate

20.01.05 Lecture 2, CS5270 9

Distributed Architecture

20.01.05 Lecture 2, CS5270 10

A Node

20.01.05 Lecture 2, CS5270 11

A Node

Often, multiple instances of the above for fault tolerance!

20.01.05 Lecture 2, CS5270 12

A Node

20.01.05 Lecture 2, CS5270 13

The Host Computer

DSP Processor

ASIC

Timer

Memory

Bus

20.01.05 Lecture 2, CS5270 14

The Host Computer

DSP Processor

ASIC

Timer

Memory

20.01.05 Lecture 2, CS5270 15

Tasks

DATA SETS

TASK1

TASK2

TASK3

TASK4

RT !mages!

20.01.05 Lecture 2, CS5270 16

RT Images

• RT entity:– Some item of interest whose value changes over

time.– Pressure, temperature, valve position …

• Continuous RT entity:– Can be observed at any point in time

pressure

• Discrete RT entity– Can be observed only between specified occurrences

of interesting events

20.01.05 Lecture 2, CS5270 17

RT Images

• RT Image:– Current picture of an RT entity.– <Name, time-of-observation, Value>

• Accuracy:– Value– Temporal

• <N, t, v> is -accurate if the value of N was v at some time in the interval (t-, t).

20.01.05 Lecture 2, CS5270 18

RT Images

• Suppose <N, v> is observed at time t and used at time t’.

• Then the maximum error (v’ – v) depends on the temporal accuracy () and the maximum gradient of N during this interval.

• If the gradient is high then must be small and tasks using N must be scheduled often! (this is a fair but crude statement)

20.01.05 Lecture 2, CS5270 19

Accuracy

RT Image Max.Change V-Accuracy T-accuracyPiston Position

6000 RPM 0.1 degrees 3secs

Acc. pedal 100%/sec 1% 10 msecs

Eng. Load 50%/sec 1% 20 msecs

Oil temp 10%/min 1% 6 seconds

20.01.05 Lecture 2, CS5270 20

The Design Challenge

• Derive a model of the closed system (external).– Specification/requirements– Timing– Notion of physical time

• Design and implement –a distributed, fault-tolerant, optimal- real time computing system so that the closed system meets the specification/requirements.

20.01.05 Lecture 2, CS5270 21

The Structural Elements

• Each computing node will be assigned a set of tasks to perform the intended functions.

• Task :– Execution of a (simple) sequential program.

Read the input data The internal state of the task (include RT profiles) Terminate with production of results and updating

internal state of the task.

20.01.05 Lecture 2, CS5270 22

Tasks

• The (real time) operating system provides the control signal for each initiation of the task.

• Stateless task: no internal state at the time of initiation.

• Task with state

20.01.05 Lecture 2, CS5270 23

Tasks

• Simple task:– No synchronization point within the task.– Does not block due to lack of progress by other tasks

in the system.– But can get interrupted (preempted) by the operating

system.– Total execution time can be computed in isolation.– The Worst Case Execution Time of task over all

possible relevant inputs. Correct estimate of WCET is crucial for guaranteeing real

time constraints will be met.

20.01.05 Lecture 2, CS5270 24

Complex Tasks• Contains blocking synchronization statement:

– “wait” semaphore operation.– “receive” message operation.

• Must wait till another task has updated a common data structure:– Data dependency– Sharing

• Must wait for input to arrive.• WCET of a complex task can not be computed

in isolation..

20.01.05 Lecture 2, CS5270 25

Interfaces

• Interfaces:– Common boundary between two subsystems.– Design is essentially interface design.– Designing and implementing the interface

“glue logic” consumes the major portion of the design cycle.

20.01.05 Lecture 2, CS5270 26

20.01.05 Lecture 2, CS5270 27

Interfaces

• Interface Parameters:– Control signals flowing across the interface

and the associated task invocations.– Temporal properties to be satisfied by the

control signals and data values flowing across.

– Functional relationships between input and output data.

20.01.05 Lecture 2, CS5270 28

The Host Computer

DSP Processor

ASIC

Timer

Memory

20.01.05 Lecture 2, CS5270 29

Tasks

DATA SETS

TASK1

TASK2

TASK3

TASK4

20.01.05 Lecture 2, CS5270 30

Tasks• There will be tasks that are triggered by

exceptions, interrupts and alarms.• There will be tasks that need to be executed

periodically.• These tasks may have precedence

relationships.• These tasks may have deadlines.• These tasks may share data structures.• They may have to execute on the same

processor.• We must schedule!

20.01.05 Lecture 2, CS5270 31

Scheduling: Basic Concepts

• Scheduling Policy: – CPU has to execute –sequentially- a set of concurrent

tasks. If T1 and T2 are both executable at t we must choose

between T! and T2

• Scheduling Algorithm: – The recipe (algorithm) which determines at each time

t which task to execute .• Dispatching:

Allocating the CPU to the task selected by the scheduling algorithm.

20.01.05 Lecture 2, CS5270 32

Scheduling: Basic Concepts

• Active Task:– A task which can potentially execute on the

CPU (which may or may not be available).• Ready Task:

– An active task which is waiting for the CPU• Running Task:

– An active task in execution.• Ready Queue:

– The queue in which ready tasks are kept.

20.01.05 Lecture 2, CS5270 33

Scheduling : Basic Concepts• Preemption:

– Tasks may be activated dynamically time of activation not determined.

– If the task activated at time is more important (has higher priority) than the running task: running task is interrupted and inserted in the running

queue. • Preemption is needed for :

– Exception-handling tasks.– Tasks may have different levels of criticality.– Improve system responsiveness, throughput,

utilization etc.

20.01.05 Lecture 2, CS5270 34

The Ready Queue

20.01.05 Lecture 2, CS5270 35

Schedules• Task set J = { J1, J2, …, Jn}• A schedule assigns at each t one task to the processor

so that each task is eventually completed.• A schedule can be preemptive.• Schedules will have to perform context switching.• A schedule is feasible if all tasks can be completed while

satisfying the given constraints.• A task set is schedulable If there is at least one

scheduling algorithm which produces a feasible schedule.

20.01.05 Lecture 2, CS5270 36

Schedule

20.01.05 Lecture 2, CS5270 37

Schedule

0 1 3 5 9

1.5? 7?

20.01.05 Lecture 2, CS5270 38

Schedule with Preemption

39

Schedule with Preemption

0 1 3 4 6 7.5 9.5

2? 5?

Context switch?

20.01.05 Lecture 2, CS5270 40

Task Constraints

• Timing Constraints• Precedence Constraints• Resource Constraints

20.01.05 Lecture 2, CS5270 41

Timing Constraints

• Timing Constraints:– A task should meet its deadline.

Hard Soft

• Relevant Parameters for the task Ji:– arrival time ai

Request time, release time

– computation time Ci

time needed to execute Ji (without interruption).

20.01.05 Lecture 2, CS5270 42

Timing Parameters

• deadline di

– time before which Ji must be completed.

• start time si

• finishing time fi

• value (priority?) vi:– The relative importance of Ji.

20.01.05 Lecture 2, CS5270 43

Basic Timing Parameters

20.01.05 Lecture 2, CS5270 44

Basic Timing Parameters

Di = di – ai

Relative deadline

20.01.05 Lecture 2, CS5270 45

Timing Parameters• Pattern of activation:

– Periodic task regularly activated at a constant rate.

instances or jobs corresponding to the same task. i the phase of I :

The activation time of the first instance of the periodic task i. Ti the period of the task. Di the relative deadline Often, one assumes Di = Ti

– Aperiodic task: Same as periodic tasks but the activation times are NOT

periodic.

20.01.05 Lecture 2, CS5270 46

Periodic/Aperiodic Task

20.01.05 Lecture 2, CS5270 47

Task Constraints

• Timing Constraints• Precedence Constraints• Resource Constraints

20.01.05 Lecture 2, CS5270 48

Precedence Constraints

• Precedence Constraints:– Tasks can not be executed in any arbitrary

order. Data dependencies. Control strategy

• Task Graphs:– Instead of Task sets.– Nodes are tasks– Edges capture precedence.

20.01.05 Lecture 2, CS5270 49

Task Graph

20.01.05 Lecture 2, CS5270 50

The Task Graph

20.01.05 Lecture 2, CS5270 51

Task Constraints

• Timing Constraints• Precedence Constraints• Resource Constraints

20.01.05 Lecture 2, CS5270 52

Resource Constraints• resource:

– software structure used by a task during its execution.– A data structure, variables, an area of main memory,

a file, a piece of code, a set of registers of a peripheral device.

• Shared resource:– Used by more than one task.

• Exclusive resource:– No simultaneous access.– Require mutual exclusion.– Operating must provide a synchronization mechanism

to ensure sequential access..

20.01.05 Lecture 2, CS5270 53

Critical Section

• Critical section:– A piece of code belonging to task executed

under mutual exclusion constraints.• Mutual exclusion enforced by

semaphores.– wait(s)

Blocked if s = 0. – signal(s)

s is set to 1 when signal(s) executes.

20.01.05 Lecture 2, CS5270 54

Structure of Critical Sections.

20.01.05 Lecture 2, CS5270 55

Wait State• A task waiting for an exclusive resource is blocked on

that resource.• Tasks blocked on the same resource are kept in a wait

queue associated with the semaphore protecting the resource.

• A task in the running state executing wait(s) on a locked semaphore (s = 0) enters the waiting state.

• When a task currently using the resource executes signal(s), the semaphore is released.

• When a task leaves its waiting state (because the semaphore has been released) it goes into the ready state:– Why not enter the running state?

20.01.05 Lecture 2, CS5270 56

Waiting State

20.01.05 57

Blocking via Exclusive Resource

J1 has higher priority than J2.

Preemption is in play.

Only one processor available.

20.01.05 Lecture 2, CS5270 58

Multiprocessor Settings

a1 e1 d H

a2 e2 s rec

20.01.05 Lecture 2, CS5270 59

Scheduling Problem.

• Task set {J1, J2,..,Jn}

• Processors {P1, P2,…, Pm}

• Resources {R1, R2, …,Rs}• Timing constraints• Precedence constraints• Resource constraints• Problem: Assign processors and resources to

tasks so that all the tasks can be finished under the imposed constraints.

20.01.05 Lecture 2, CS5270 60

Scheduling Problem

• The general problem (in fact various simpler versions of it) is NP-complete.

20.01.05 Lecture 2, CS5270 61

Scheduling Problem• The general problem (in fact various simpler versions of

it) is NP-complete.• There is a non-deterministic Turing Machine TM and a

polynomial in one variable p(n) (egs. 8n3 + 5n + 6) such for each problem instance of size n (in binary representation!), TM determines if there exists a schedule and if so outputs one in atmost p(n) steps.

• Any non-deterministic polynomial time problem can be transformed in deterministic polynomial time to the general scheduling problem.

• Only exponential time deterministic algorithms are known.

20.01.05 Lecture 2, CS5270 62

Scheduling Problem

• Algorithm1 O(n)• Algorithm2 O(5n)• Each computation step 1 sec.• n = 30• Algorithm1 : 30 seconds.• Algorithm2 : 30, 000, 000 years!

20.01.05 Lecture 2, CS5270 63

Scheduling Problems• Must find imperfect but efficient solutions to

scheduling problems.• Great variety of algorithms exist:

– various assumptions– Different complexities– Different pragmatic contents.

• Optimal scheduling algorithm:– Minimizes a given cost function.– If no cost function, then no algorithm in the same

class can produce a feasible schedule if the optimal one can not.

20.01.05 Lecture 2, CS5270 64

A Classic Example

• Rate Monotonic Scheduling.– Task set : {J1, J2, …, Jn}– Each task is periodic. T1, T2,.., Tn

– i = 0 for each i.– Di = Ti for each i.– Pre-emption allowed.– Only one processor– No precedence constraints– No shared resources.

20.01.05 Lecture 2, CS5270 65

RMS

• The RMS algorithm:– Assign a static priority to the tasks according

to their periods. Tasks with shorter periods have higher priorities.

– Preemption policy: If Ti is executing and Tj arrives which has higher

priority (shorter period), then preempt Ti and start executing Tj.

20.01.05 Lecture 2, CS5270 66

RMS Results

• RMS is optimal.– If a set of of periodic tasks (satisfying the

assumptions set out previously) is not schedulable under RMS then no static priority algorithm can schedule this set of tasks.

• RMS requires very little run time processing.

• Static scheduling policy.

20.01.05 Lecture 2, CS5270 67

Process Utilization Factor• Task set = {T1, T2, …, Tn}• Process Utilization Factor

– Ci / Ti

– C1 / T1 + C2 / T2 + … Cn / Tn

• If this factor is GREATER than 1 then the task set can not be scheduled.– Why?

• If UF ≤ 1 it may be schedulable.• If UF Ulub then it is guaranteed to be

schedulable.

20.01.05 Lecture 2, CS5270 68

Process Utilization Factor

• Task set = {T1, T2, …, Tn}• If UF Ulub then it is guaranteed to be

schedulable.• Ulub = n( 21/n – 1)• For large n this is approximately 0.69.• But if UF is greater than Ulub and not greater

than 1, we must check explicitly whether the task set is schedulable (under RM).

20.01.05 Lecture 2, CS5270 69

EDF

• Earliest Deadline First.– Tasks with earlier deadlines will have higher

priorities.– Applies to both periodic and aperiodic tasks.– EDF is optimal for dynamic priority algorithms.– A set of periodic tasks is schedulable with

EDF iff the utilization factor is not greater than 1.

20.01.05 Lecture 2, CS5270 70

An Example

• {T1, T2}• T1

– Period = 5 – Computation time = 2

20.01.05 Lecture 2, CS5270 71

An Example

• {T1, T2}• T2

– Period = 7 – Computation time = 4

20.01.05 Lecture 2, CS5270 72

An RMS Schedule ?

20.01.05 Lecture 2, CS5270 73

An RMS Schedule

Time-Overflow

20.01.05 Lecture 2, CS5270 74

The Example

• UF = 2 / 5 + 4 / 7 = 0.4 + 0.57 = 0.97• Guaranteed to be schedulable under EDF!

20.01.05 Lecture 2, CS5270 75

Resource Access Protocols

• Multiple tasks.• Uniprocessor• Shared resources.

– Need proper protocols for accessing shared resources.

– Resource access protocols.• Avoid priority inversion!