What Shape is Your Real-Time System?

Lonnie R. WelchIntelligent, Real-Time, Secure Systems Lab.

School of EECSOhio Universitywelch@ohio.edu

Michael W. MastersNaval Surface Warfare

CenterDahlgren Division

MastersMW@nswc.navy.mil

Overview

• Patterns in real-time mission-critical systems

• Basis for a taxonomy of RTMCSs

• Useful for – technology selection

– identifying research needs

Ship Computational Resource Pool

OpticalOpticalCrosslinkCrosslink

PassivePassiveOpticalOptical

Ka CrosslinkKa Crosslink

In-situ UserIn-situ UserPC Based GSPC Based GSIn-situ UserIn-situ UserPC Based GSPC Based GS

Comm Comm GatewayGatewayComm Comm

GatewayGateway MetadataMetadataWarehouseWarehouseMetadataMetadata

WarehouseWarehouse

NASA GSFC Earth Science Vision-Distributed Information-System-in-the-

SkyCommercialCommercial

Communication Communication NetworkNetwork

Active Active OpticalOptical

DigitalDigitalLibraryLibraryDigitalDigitalLibraryLibrary

KaKa

Optical CrosslinkOptical Crosslink

KaKa

KaKa

• Interoperating Interoperating Measurement Measurement Systems (Air / Systems (Air / Spacecraft / In-situ)Spacecraft / In-situ)

• Flexible Flexible Measurement Measurement Network Network ArchitectureArchitecture

• Direct Distribution Direct Distribution of Derived Productsof Derived Products

• Network Network Computing-in-the-Computing-in-the-SkySky

System Agility is Needed

unpredictable environments (e.g., war-fighting situations)

system intrusions

harsh conditions (resulting in damage)

Most Previous Work

“worst-case” execution time (WCET) known a priori for each job [Liu73, Ram89, Xu90, Sha91, Bak91]

static resource allocation and guarantees; low agility

poor resource utilization when WCET normal execution time [Ram89, Leh96, Hab90]

accurately measuring WCET is often difficult, and sometimes impossible [Ste97, Abe98]

Adaptive QoS Management

Diagnose

Monitor QoS violation(s)

Causes and possible actions

for recovery

1 2

Analyze

3

Allocate

4

“Best” recovery actions

Start an RT system in a feasible allocation

Maintain a feasible allocation

Real-Time SystemDeveloper

Start the RM executables

Real-Time System

Installer

Initiator

Initiator

Initiator

Host

Operator

Critical Use Cases

Analysis Packages

Instrumentation and Control ManagementInstrumentation and Control Management

Allocation ManagementAllocation

Management

Real-Time System Management

Real-Time System Management

User Management

User Management

Specification File Management

Specification File Management

Inter-Class Collaborations

Real-Time System

:Real-Time SystemInterface

1: Real-Time SystemEvent

:Real-Time SystemEvent History

2: StoreEvent

:Real-Time SystemMonitor

5: Get

:Real-Time SystemState

6: Update

4: Get

:Host-ResourceMonitor Interface

:AllocationManager

11: AllocateResource

14: Get 16: Set

:Real-Time SystemDiagnosis

7: Diagnose

9: Get

8: Get

17: Update

Host

:AllocationState

:Real-Time SystemInformation

10: Get

:ResourceInformation(hardware)

13: Get

12: Get

:ResourceMonitor

:AllocationController

15: AllocateResource

3: ComputePerformance

“Maintain a FeasibleAllocation” use case

• Air traffic control

• Satellite C2

• Air defense

• Squads of mobile robots, UAVs, satellites

environment

sensors

actuators

assessment

control

initiationevent

event

DataSource

DataHandler

DataStream

Situation Assessment Use Case

• sensor periodically produces a stream of data elements

• decide if actions should be performed

• variable data set size

• heterogeneous elements

λobs < λreq

EventSource

EventHandler

Event Stream

Action Initiation Use Case

λobs < λreq

• plan and initiate an action to handle event

• environment-dependent event arrival rate

• heterogeneous events

• timely performance is mandatory

Control Use Case

DataSource Event-

DrivenPeriodic

DataHandler

DataStream

EventSource

Event Stream

• activated by an event

• guides actions to success

• deactivated by an event – the completion of the action

• period deadline and action completion deadline

• completion deadline is dynamic

control

initiationassessment

<<generalize>><<generalize>>

Characterizing Design Patterns

• Larger granularity than task or object

• Cannot characterized accurately by worst case

• Fixed set of applications, with varying loads

• How to characterize the problem space?

Taxonomy Categories

• Properties of a real-time system

• Properties of the environment (which effect the real-time system)

• The set of properties defines a “shape”

Properties of a Real-Time System

PatternBehavior

Timing Requirement

TaskRelations

Forms ofAdaptation

Timing Requirement

Granularity Strictness

Abstraction levelComplexity

Properties of the Environment

Dynamics

CharacteristicsWorkload

Properties of Workload

event arrival rate

data stream size

periodstream elements

Uses of the Taxonomy

• Characterizing applications

• Categorizing real-time technology

• Selecting appropriate techniques for engineering of a particular system

• Identifying open research areas

Characterizing RT (sub)Systems

• The situation assessment path in an air defense system

• Periodically reviews all radar tracks

• If a threat track is detected, notifies the missile engagement path

Situation Assessment

Design Pattern Behavior

Timing Requirement

TaskRelations

Forms ofAdaptation

assessmentinitiation

guidance

PeriodicTransient

Transient-periodic

Independent

Dependent

Rrsc. alloc.

Precision

Concurrency

Slack

Situation Assessment

Granularity Strictness

Abstraction level

Complexity

Soft

Importance

Utility

Hybrid

Hard Firm

Multiple

Single

msec

sec

minutes

hours

taskobject

methodinstr

taskgroup

Situation Assessment

Dynamics

Characteristics

Workload

static

time invariant stochastic

dynamic

time variant stochastic

hybrid

burtsy

constant

gradually changing

Situation Assessment

event arrival rate

data stream size

period

stream elements

constant

set

interval

distribution

dynamic

constantset

interval

distribution

dynamic

homogeneousset

interval

distribution

unknown

fixed

set

interval

unconstrained

Categorizing Technology• RMA - the rate monotonic technique for

schedulability analysis

• Determines if a set of periodic, independent tasks is schedulable

• The worst case execution time is known for each task

Rate Monotonic Analysis

Design Pattern Behavior

Timing Requirement

TaskRelations

Forms ofAdaptation

assessmentinitiation

guidance

PeriodicTransient

Transient-periodic

Independent

Dependent

Rrsc. alloc.

Precision

Concurrency

Slack

Rate Monotonic Analysis

Granularity Strictness

Abstraction level

Complexity

Soft

Importance

Utility

Hybrid

Hard Firm

Multiple

Single

msec

sec

minutes

hours

taskobject

methodinstr

taskgroup

Rate Monotonic Analysis

Dynamics

Characteristics

Workload

static

time invariant stochastic

dynamic

time variant stochastic

hybrid

burtsy

constant

gradually changing

Rate Monotonic Analysis

event arrival rate

data stream size

period

stream elements

constant

set

interval

distribution

dynamic

constantset

interval

distribution

dynamic

homogeneousset

interval

distribution

unknown

fixed

set

interval

unconstrained

Technology Selection• Is the application region contained within the

space covered by the technology?

• Inefficiencies may result if the technology space is larger than the application region

• Errors may result if the technology space does not contain the application region

Identifying Research Needs

• Define the shapes of existing technology

• Which shapes are missing?

• Who cares?

• Engineers of applications not having corresponding technologies