DoD Perspective on the Future of Embedded Software Developmentportals.omg.org/hpec/files/hpec-si/keynote.pdf · DoD Perspective on the Future of Embedded Software Development. 2

Dr. Charles J. HollandDr. Charles J. HollandDirector, Information SystemsDirector, Information Systems

Deputy Under Secretary of Defense (Science & Technology)Deputy Under Secretary of Defense (Science & Technology)

DoD Perspective on the Future ofEmbedded Software Development

2

Joint Vision 2020(Key Enabler - Information Superiority)

The joint force of 2020 will use superior information andknowledge to achieve decision superiority in order to reach fullspectrum dominance, achieved through the interdependentapplication of:

The Lenses of Technological Innovation and Information SuperiorityIntegrate and Amplify the Four Operational Concepts

,QIRUP

DWLRQ6XSHULRULW\

)RFXVHG /RJLVWLFV

3UHFLVLRQ (QJDJHPHQW

'RPLQDQW 0DQHXYHU

)XOO 'LPHQVLRQ

DO 3URWHFWLRQ

,QQRYDWLRQ

)XOO

6SHFWUXP

'RPLQDQFH

3

Changing Strategic Environment

Global US InterestsPolitical - Economic - Humanitarian

Globalization of Technology

Asymmetric ThreatsIn any domain - Air, Land, Sea, Space or Information

SECRETARY OFDEFENSE

SECRETARY OFDEFENSE

Acquisition

DASD, INTELLIGENCEMr. Christopher K. Mellon (Actg)

3E187 695-7181

DASD, SECURITY & INFORMATIONOPERATIONS

Mr. William Leonard (Actg)3E194 695-2396

DASD, C3I, SURVEILLANCE,RECONNAISSANCE

& SPACERADM Robert Nutwell

3D174 697-8613

DEP DASD, C3I CIO Mr. Paul Brubaker (Actg)

3E243 695-0871

DASD, PROGRAMS & EVALUATION

Ms. Cheryl J. Robey3E240 697-3426

1 March 2000

OFFICE OF ECONOMIC

ADJUSTMENTMr. Paul J. Dempsey400 AND 604-6020

DIR, STRATEGIC & TACTICAL SYSTEMSDr. George R. Schneiter

3E130 697-9386

DIR, SMALL & DISADVANTAGED

BUSINESS UTILIZATIONMr. Robert L. Neal, Jr.

RPN Suite 9100 588-8620

DUSD, INSTALLATIONSMr. Randall A. Yim 3E1074 697-1771

DUSD, INDUSTRIAL AFFAIRS

Mr. Jeffrey P. Bialos 3E1060 697-0051

DUSD, ENVIRONMENTAL SECURITY

Ms. Sherri W. Goodman3E792 695-6639

DUSD, ACQUISITION REFORM

Mr. Stan Z. Soloway3E1034 695-6413

DIR, DEFENSE PROCUREMENT

(Vacant)3E1044 695-7145

DIR, ACQUISITIONRESOURCES & ANALYSIS

Dr. Nancy L. Spruill3E1025 614-5737

DIR, INTERNATIONAL COOPERATION

Mr. Alfred G. Volkman 3A280 697-4172

DIR, INTEROPERABILITY Dr. V. Garber

3E144 695-9713

DUSD, ADVANCED SYSTEMS & CONCEPTSMr. Joseph J. Eash, III3E1014 697-6446

CIO & IntelMatters

EXEC DIR, DEFENSE SCIENCE BOARDMr. John V. Ello

3D865 695-4157

DIR, SPECIAL PROGRAMSMGen H. Marshal Ward,

USAF 3D1064 697-1282

DIR, OPERATIONAL TEST & EVALUATION

Honorable Philip E. Coyle3A1073 697-3654

DIR, ADMINISTRATIONMr. Larry S. Barlow3D1020 697-2525

DIR, DEFENSE SECURITY COOPERATION AGENCY

LTG Michael S. Davison, USACGN 303 604-6604

ASSISTANT SECRETARY OF DEFENSE (COMMAND,CONTROL, COMMUNICATIONS & INTELLIGENCE)

Honorable Arthur L. Money

Principal Deputy: Dr. Linton Wells II3E172 695-0348

ASSISTANT SECRETARY OF DEFENSE (COMMAND,CONTROL, COMMUNICATIONS & INTELLIGENCE)

Honorable Arthur L. Money

Principal Deputy: Dr. Linton Wells II3E172 695-0348

DUSD, LOGISTICSMr. Roger W. Kallock

3E114 697-5531

DIR, DEFENSE LOGISTICS AGENCY

LTG Henry T. Glisson, USAFT BEL 767-5200

DIR, DEFENSE ADVANCEDRESEARCH PROJECTS

AGENCYDr. Frank L. Fernandez

VSP 983 696-2400

DIR, TECHNOLOGY SECURITY

Mr. Dave Tarbell 400 AND 604-5215

DIR, BALLISTIC MISSILE DEFENSE ORGANIZATIONLt Gen Ronald T. Kadish,

USAF1E1081 693-3025

ATSD, NUCLEAR, CHEMICAL AND

BIOLOGICAL DEFENSE PROGRAMS

(Vacant)

AIR FORCE SAEHonorable Lawrence Delaney

4E964 697-6361

ARMY SAEHonorable Page Hoeper

2E672 695-6153

NAVY SAEHonorable H. Buchanan

4E741 695-6315

DIR, DEFENSE THREAT REDUCTION AGENCY

Dr. Jay DavisDULLES 810-4883

DEPUTY UNDER SECRETARY OF DEFENSE(ACQUISITION & TECHNOLOGY)

Honorable David R. Oliver 3E1006 697-7021

DEPUTY UNDER SECRETARY OF DEFENSE(ACQUISITION & TECHNOLOGY)

Honorable David R. Oliver 3E1006 697-7021

DUSD, SCIENCE & TECHNOLOGY

Dr. Delores M. Etter (Des)3E808 695-0598

DIRECTOR DEFENSE RESEARCH & ENGINEERING

Honorable Hans Mark 3E808 697-5776

DIRECTOR DEFENSE RESEARCH & ENGINEERING

Honorable Hans Mark 3E808 697-5776

UNDER SECRETARY OF DEFENSE(ACQUISITION, TECHNOLOGY AND LOGISTICS)

Honorable J. S. Gansler693-4265 3E933

_____________________PRINCIPAL DEPUTY

Honorable David R. Oliver, Jr.697-7021 3E1006

UNDER SECRETARY OF DEFENSE(ACQUISITION, TECHNOLOGY AND LOGISTICS)

Honorable J. S. Gansler693-4265 3E933

_____________________PRINCIPAL DEPUTY

Honorable David R. Oliver, Jr.697-7021 3E1006

DEPUTY UNDER SECRETARY OF DEFENSE (LOGISTICS & MATERIEL READINESS)

(Vacant) 3E114 697-5531

DEPUTY UNDER SECRETARY OF DEFENSE (LOGISTICS & MATERIEL READINESS)

(Vacant) 3E114 697-5531

6

Director, DefenseResearch & Engineering

NUCLEAR TREATY PROGRAMDr. Ralph Alewine, III

DIRECTOR, DEFENSE RESEARCH AND ENGINEERING

Honorable Hans Mark

DUSD, ADVANCED SYTEMS & CONCEPTS

Mr. Joseph J. Eash, III

DIR, STRATEGIC & TACTICAL SYSTEMS Dr. George R. Schneiter

DUSD, SCIENCE & TECHNOLOGYDr. Delores M. Etter

DIR, DEFENSE THREAT REDUCTION AGENCY

Dr. Jay C. Davis

DIR, DEFENSE ADVANCED RESEARCH PROJECTS AGENCY

Dr. Frank L. Fernandez

ODASD, NUCLEAR MATTERSDr. Frederick Celec

CHEMICAL AND BIOLOGICALDEFENSE PROGRAMS

Dr. Anna Johnson-Winegar

DIR, BALLISTIC MISSILE DEFENSEORGANIZATION

LGen Ronald Kadish, USAF

7

Deputy Under Secretary ofDefense (Science and Technology)

Open Systems JointTask Force

Dir: Mr. Tom Smith

Technology TransferDir: Mr. John Todaro

Defense Modeling &Simulation Office

Dir: Col Forrest Crain

Jan 3, 2000

S&T Plans and ProgramsS&T Laboratories

Dir: Mr. Robert Tuohy

International Plans& Programs

Dir: Mr. Barry DeRoze

Defense TechnologyAnalysis Office

Dir: Mr. James Whitt

S&T Planning S&T Planning S&T ProgramsS&T Programs

Information SystemsDir: Dr. Charles Holland

HPCMODir: Mr. Cray Henry

Multidisciplinary SystemsDir: Col Al Shaffer

Sensor SystemsDir: Dr. Jasper Lupo

Weapons SystemsDir: Dr. George Ullrich

Basic ResearchDir: Dr. Robert Trew

Bio Systems Dir: Dr. Bob Foster

SERDPDir: Mr. Brad Smith

Deputy Under Secretary of Defense(Science and Technology)

Dr. Delores M. Etter

Executive CommitteeDr. George Ullrich

Dr. Charles HollandMr. Robert Tuohy

CAPT Dave Gillard

Software IntensiveSystems

Dir: Dr. Jack Ferguson

S&T TransitionS&T Transition

Administrative OfficeMilitary Assistant: CAPT Dave GillardConfidential Assistant: Mrs. Pat HallSpecial Assistants: Ms. Laura Douglass

Ms. Mary HutchinsMr. Alonzo Davis

8

S&T & DT&E

Applications

S&T S&T

& DT&E& DT&E

ApplicationsApplications

CHSSI for CTAsCHSSI for CTAsCHSSI for CTAsSoftware Support InitiativeSoftware Support InitiativeSoftware Support Initiative

CommercialOff-The-Shelf Software Commercial CommercialOff-The-Shelf SoftwareOff-The-Shelf Software

ProgrammingEnvironment & Training Programming ProgrammingEnvironment & TrainingEnvironment & Training

Support ServicesSupport ServicesSupport Services

Computer Systems8 Vector8 Scalable Parallel8 Clustered Workstations

Computer Computer Systems Systems88 Vector Vector88 Scalable Parallel Scalable Parallel88 Clustered Clustered Workstations Workstations

Storage8 Distributed File Systems8 Robotics Archival System

Storage Storage88 Distributed File Distributed File Systems Systems88 Robotics Archival Robotics Archival System System

ScientificVisualization Scientific ScientificVisualizationVisualization

Internal NetworkingInternal NetworkingInternal Networking

Wide Area ConnectivityWide Area ConnectivityWide Area Connectivity

NetworkingNetworkingNetworking

Distributed SIP/RT HPC Centers AFRL/IF, AFRL/SN, SSCSD, WSMR, NAWC, AEDC, MHPCC, RTTC, TARDEC, AAC

Distributed SIP/RT HPC CentersDistributed SIP/RT HPC Centers AFRL/IF, AFRL/SN, SSCSD, AFRL/IF, AFRL/SN, SSCSD, WSMR, NAWC, AEDC, MHPCC, WSMR, NAWC, AEDC, MHPCC, RTTC, TARDEC, AAC RTTC, TARDEC, AAC

DoD HPC Modernization Program DoD HPC Modernization Program DoD HPC Modernization Program

Signal/Image Processing CTA Radar, Sonar, SAR, ATR, IR and Hyperspectral Image Exploitation

Signal/Image Processing CTASignal/Image Processing CTA Radar, Sonar, SAR, ATR, Radar, Sonar, SAR, ATR, IR and Hyperspectral Image IR and Hyperspectral Image Exploitation Exploitation

HPCMP Investments in Real-Time Embedded HPC

Integrated Modeling and Test CTA Tracking, Image Classification, RT Model Validation, Non-Uniformity Correction

Integrated Modeling and Test CTAIntegrated Modeling and Test CTA Tracking, Image Classification, RT Model Tracking, Image Classification, RT Model Validation, Non-Uniformity Correction Validation, Non-Uniformity Correction

PET Activities Yearly SIP Forum, VSIPL Tiger team RTExpress SIP Repository

PET ActivitiesPET Activities Yearly SIP Forum, Yearly SIP Forum, VSIPL Tiger team VSIPL Tiger team RTExpressRTExpress SIP RepositorySIP Repository

9

Outline

• The Challenge of High-Performance Embedded Computing

• The Current State of High-Performance Computing

• A Vision for the Future

• Summary

10

Embedded Processing Spectrum

• Broad range of applications with very different stressing requirements

• All possess similar software development concerns

Platform Size (liters) Weight (kg) Power (W) Processing

(GFLOPS)

Large radar/surveillance platforms (JSTARS, AWACS, AEGIS, THAAD)

1000's 500-1000 10-20 kW 1,000 – 10,000

Unmanned surveillance platforms (UCAV, Global Hawk, Discoverer II, etc.)

30 – 100 50 – 100 100–500 W 50 - 1,000

Use-once assets (MALD, torpedos, missiles) 1 – 30 1 – 50 ~100 W 1 – 100

Micro-assets (micro-UAVs, small unit operations, smart dust)

0.01 – 0.1 0.1 – 1 ~0 – 10 W 0.1 – 10

Processor Estimates

11

Evolution towards HPEC

EmbeddedComputing

• Embedded processors movingtowards high performance

• Major drivers:– Latency, throughput, determinism

– Form factor

EHPC

HPEC

HPEC Applications

• Market providing more hardware choices for embedded applications

• Software standards must evolve to support any hardware platform

High PerformanceComputing

• HPC evolving into embeddedapplications

• Major drivers:– Throughput

– Ease of development

12

The Challenge:Software Interoperability

Standards-basedSoftware

Interoperability

• Move towards “write once, run anywhere”– Allows for joint development on workstation class machines

– Provides easy porting on new machines, allowing for “technology refresh”

• Productivity, portability and scalability without sacrificing performance– Productivity: Easier code development and maintenance

– Portability: Permits migration across platforms

– Scalability: Allows for easy growth into evolving requirements

• Move towards “write once, run anywhere”– Allows for joint development on workstation class machines

– Provides easy porting on new machines, allowing for “technology refresh”

• Productivity, portability and scalability without sacrificing performance– Productivity: Easier code development and maintenance

– Portability: Permits migration across platforms

– Scalability: Allows for easy growth into evolving requirements

HPC and Networked Computers Embedded Computing Platforms

13

Model-Year Portability

Increasing Network Performance(latency, throughput, bisection bandwidth, ...)

Incr

easi

ng

Pro

cess

ing

Cap

abili

ty(g

ener

atio

ns)

Increasing overall system performance

Processor support limitedor dropped in later products

Vendor A product line

Vendor B product line

Portable software leverages inevitable advances innetwork and processor technology

“Point” solutions specific to a single vendor’shardware or software are long-term cost ineffective

Portable software with highperformance is the best solution

Portable software with highperformance is the best solution

System Development/Acquisition4 Years 4 Years 4 Years

Milestones

TechnologyDevelopment

FieldDemonstration

Engineering/ManufacturingDevelopment

Insertion

SignalProcessor Gen 1 Gen 2 Gen 3 Gen 4 Gen 6Gen 5

…military development and deployment cycles are long.

…military development and deployment cycles are long.

COTS HPEC system lifetimes are short…COTS HPEC system lifetimes are short…

14

June 9, 1999 "Defense Science and Technology: Preparing for the Future,”Presentation to DARPA; Dr. Delores M. Etter, DUSD (Science & Technology)

Military Software Complexity

• Moore’s Law addressescomputation, not complexity!

• In 1995, 85% of military softwareprojects finished over time and/orbudget

• 1/2 of projects double costestimates

• Projects slip an average of 36months

• 1/3 of projects cancelled due toschedule/cost slips

Major Defense Systems

InformationSystems

Scientific and Engineering

PersonalApplications

CommercialPackages

100 1,000 100,000 1,000,00010,000

100

1,00

010

,000

Number of Function Points

Co

st p

er F

un

ctio

n P

oin

t ($

)

Increased Cost

and Complexity

denotes average

• Software complexity of signal and image processing applications

– System development overly complex already

– Cannot let sensor processing compound the problem

• Software complexity of signal and image processing applications

– System development overly complex already

– Cannot let sensor processing compound the problem

15

Outline




• Summary

16

Parallel Architectures andOperating Environments

HP

EC

HP

C

Architecture ProgrammingM odel

M iddleware Ease ofUse

M emoryI/O Overlap

Real-TimeIssues

SymmetricMultiprocessor

Shared Memory /Single System Image

Threads,Compiler Pragmas

Easy Threads Non RT-OS

Hybrid Shared Memory /Single System Image

Threads,Compiler Pragmas

Easy Threads Non RT-OS,Page Migration

Multicomputer Message passing Threads Non RT-OS

Multicomputer Distributed MemoryMessage Passing

One-sided DMA

Message passing

Hard

DMA engines(mainly)

RT-OS

For most applications, achieving good scalability requiresexpert architecture knowledge

High-level middleware standards can simplify this!

• Provide abstraction for application-level functions- Hide low-level mechanisms for parallel processing / communication

• Unify programming model across architectures- Eliminate ease-of-use issues between shared / distributed memory

- Choose hardware based on mission needs alone

High-level middleware standards can simplify this!

• Provide abstraction for application-level functions- Hide low-level mechanisms for parallel processing / communication

• Unify programming model across architectures- Eliminate ease-of-use issues between shared / distributed memory

- Choose hardware based on mission needs alone

17

Roles of Current Standard Libraries

P0 P1 P2 P3

NodeController

Parallel Signal Processor

SystemController

ConsolesOther

Computers

Low BandwidthControl Communication:

CORBA

High Bandwidth Data Communication:

MPI, MPI/RT

Single Processor Computation:

VSIPL

In development:• High-performance

CORBA Extensions• Data Re-org

In development:• High-performance

CORBA Extensions• Data Re-org

• Standards in the HPEC world fill three basic roles– Control communication– Data communication– Single processor computation

• The standards are not tightly integrated

18

HPC/HPEC Middleware AssessmentComputation Libraries

HPEC HPCMiddleware VSIPL Standard

Vendor Proprietary

BLAS / LAPACK StandardATLAS / FFTW

Vendor Proprietary

StandardsSupport

EmergingUser Acceptance TBD

WidespreadBroad User Acceptance

Object Oriented VSIPL Object-Based R&D

Automatic Tuning None R&D

Signal Processing VSIPL API No Industry Standard

Full LinearAlgebra

VSIPL API DefinedEmerging Implementations

BLAS & LAPACK

OverallAssessment

• HPEC: follow HPC lead (common interface mentality)• Full use of object-orientation can reduce API complexity• HPEC can leverage R&D in automatically tuning software• Communities should work together to develop unified

interfaces

19

HPC/HPEC Middleware AssessmentCommunication and Integrated Libraries

HPEC HPCMiddleware MPI Standard

MPI/RT StandardData Reorg (developing)

Proprietary Message Passing

MPI Standard

ScaLAPACKProprietary Message Passing

Standards Support MPI: Performance ConcernsMPI/RT: Slow Adoption

MPI: WidespreadMPI/RT: None

High-level Objects Data Reorg (Object-Based)Comm Only

HPF / ScaLAPACK (Procedural)Integrated Comm and Compute

AddressesReal-Time Concerns

MPI/RT and Data Reorg(Early Binding for PredictableCommunication Performance)

No

OverallAssessment

• Support for MPI is emerging in HPEC arena• MPI performance for HPEC needs to be evaluated further• No widely accepted standards that integrate communication

and computation

20

IntegrateComputation & Communication

Today’s Applications

CommunicationData (MPI)

CommunicationLibrary(N CPU)

ComputationData (VSIPL)

ComputationLibrary(1 CPU)

Custom

Processor Interconnect

Tomorrow’s Applications

Data Object

Computation +Communication

Library

Processors Interconnect

•Software standards developed separately•Compute & communication data are storeddifferently

•Application code necessary to pass databetween two environments

•Software standards developed separately•Compute & communication data are storeddifferently

•Application code necessary to pass databetween two environments

•Co-develop standards•Common compute &communication data format

•Seamless transitionbetween two environments

•Co-develop standards•Common compute &communication data format

•Seamless transitionbetween two environments

21

Outline




• Summary

22

Software for the Next Generation

Today’sPractice-Low level protocols-Procedural-Large platforms-Large teams-Static signal flow-Redundant hardware-Separate Comp/Comm

Tomorrow’sApplications

-Standards based-Object oriented-Small platforms

-Small teams-Dynamic flow

-Fault tolerant software-Integrated Comp/Comm

New ApplicationsRequire

New Approaches

23







Standards

•Allows portable applications

•Provides a common interface fordesigners

•Separates hardware from software

•Enables innovation in both domains

Standards

•Allows portable applications

•Provides a common interface fordesigners

•Separates hardware from software

•Enables innovation in both domains

24







Object Oriented Programming

•Essential tool for modern softwaredevelopment

•Enables code re-use by hidingcomplexity

•Allows small teams to be moreproductive

•Better compiler optimization canachieve better performance

Ready for HPEC

Object Oriented Programming

•Essential tool for modern softwaredevelopment

•Enables code re-use by hidingcomplexity

•Allows small teams to be moreproductive

•Better compiler optimization canachieve better performance

Ready for HPEC

25







Reconfigurable Software

•New applications will be morecomplex and more dynamic

•Fault tolerance will be implementedin software and hardware

•Software needs to adapt to hardware

•Employ self-optimizing softwaretechniques (e.g., FFTW & ATLAS)

Reconfigurable Software

•New applications will be morecomplex and more dynamic

•Fault tolerance will be implementedin software and hardware

•Software needs to adapt to hardware

•Employ self-optimizing softwaretechniques (e.g., FFTW & ATLAS)

26







Unify Computation & Communication

•Eliminates parallel programmingcode overhead

•Provides a clean interface fordevelopers

•Enables truly scalable programs

•Allows 3rd party parallel softwareapplications

Unify Computation & Communication

•Eliminates parallel programmingcode overhead

•Provides a clean interface fordevelopers

•Enables truly scalable programs

•Allows 3rd party parallel softwareapplications

27

Middleware as a Tool for BuildingHigh-Level Development Tools

Today Tomorrow

ProgrammableHardware

Vendor Libraries / OS

(Often) Custom Computation

Libraries

(Often) Custom Communication

Libraries

High-level tool

• Migrate away from buildingcustom infrastructure

• High-level tools need high levelsof abstraction in middleware

• Compiler challenges for automaticmapping / code generation tools

Programmable Hardware

Vendor-Specific Optimizations

High-level tool

Other Middleware(profiling, data collection, ...)

Standard Computation Libraries

Standard Communication Libraries

Object OrientedProgramming

Unified Computation & Communication

28

Outline




• Summary

29

Conclusions

• HPEC presents unique software challenges• Increasingly complex applications & constraints• Technology refresh and lifecycle support• Diverse platforms

• Standards are just beginning to emerge• HPC leads the way in software innovation• HPEC can leverage HPC approach and technology

• Vision: Write once, run anywhere• Move towards object-oriented standards• Unify computation & communication

30

Focused Mission R&DNew Ideas, Knowledge

High Risk, High Payoff Innovation, Transition

Maximum NationalSecurity Payoff

Service Labs Universities

DARPA Industries

Expanded Resource Base

InteragencyInteragency

Coalition Capability

International

DoD S&T is a Partnership

31

Backup slides

32

Software and System Costs

Command&

Control

Software LOC

SIP

Software Class Complexity PerformanceCommand and Control Difficult ModerateSignal/Image Processing Moderate Demanding

• Software lines of codedominated by command andcontrol, but...

33


Command&

Control

Software LOC

SIP


Hardware Required

C&C SIP


• Hardware costs driven by SIPfunctions

34

C&C SIPC&C SIP


Command&

Control

Software LOC

SIP


Hardware Required

C&C SIP



– Replication costs (per copy)

35

C&C SIPC&C SIP


Command&

Control

Software LOC

SIP



– Replication costs (per copy)

• Poor SIP softwareimplementation will decreasearchitecture efficiency and driveup costs


Hardware Required

C&C SIP

Signal and Image processing software drives total ownership costs – Development, hardware costs, maintenance, upgrades

Documents

DoD Perspective on the Future of Embedded Software Developmentportals.omg.org/hpec/files/hpec-si/keynote.pdf · DoD Perspective on the Future of Embedded Software Development. 2