Dr. Frederica Darema

1
Symbiotic Measurement and Simulation
Application Systems
Dr. Frederica Darema NSF
2
Outline

• Background and New Directions
• Technology Trends
• Applications and Platforms
• Symbiotic Measurement and Simulation
• Computing and Communications Systems -
  Challenges and Approaches
• Performance Engineering
• Application development and run-time support
• Agency Efforts
• Existing programs,
• Present and Future Initiatives
• Technology transfer to industry

3
Application Directions
Past

• Mostly monolithic
• Mostly one programming language

• Computation Intensive
• Batch
• Hours/days

• Computation Intensive
• Data Intensive
• Real Time
• Few Minutes/hours
• Visualization (real time)
• Interactive Steering
• Integrated ExperimentsSimulations

Present / Future

• Multi-Modular
• Multi-Language
• Multiple Developers
• Multi-source Data

4
MSTAR (DARPA)(Moving and Stationary Target
Acquisition and Recognition)
Focus of Attention
Index Database (created off-line)
...
Search Tree
Regions of Interest (ROI)
Segmented Terrain Map
SAR Image Collateral Data - DTED, DFAD - Site
Models - EOSAT imagery
...
Indexing
Target Scene Model Database (created off line)
Task Predict
Task Extract
Extract
Search
Statistical Model
Predict
Clutter Database
CAD
Match Results
Tree Clutter
Semantic Tree
Form Associations
Refine Pose Score
Analyze Mismatch
Shadow Obscuration ?
x2,y2, ??
x1,y1, ??
Score 0.75
Ground Clutter
Feature-to-Model Traceback
Match
5
Platform Directions
Past

• Vector Processors
• SIMD MPPs

• Distributed Memory MPs
• Shared Memory MPs

Present

• Distributed Platforms,
  Heterogeneous Computers and Networks
• Heterogeneity
• architecture
• (compute network)
• node power
  (supernodes, PCs)

Future

• Latencies
• variable (internode, intranode)
• Bandwidths
• different for different links
• different based on traffic

GiBs
Grids
Petaflops Platform (Grid-in-a-Box)
Distributed Platform
.
MPP
NOW
SP
6
Symbiotic Measurement Simulation Systems
USERMEASUREMENTS
USER
Dynamic Feedback Control Loop
7
Examples of Applications that can benefit from
this paradigm

• Engineering (Design and Control)
• aircraft design
• oil exploration
• computing systems hardware and software design
  (performance engineering)
• Crisis Management
• transportation systems (planning, accident
  response)
• weather, hurricanes/tornadoes, floods
• fire propagation
• Medical
• customized radiation treatment, x-rays, NMR, etc
• epidemics
• Manufacturing/Business/Finance
• Production Planning and Control
• Stock Mkt, Portfolio Analysis

8
Challenges

• Application development
• Algorithms
• tolerant to perturbations of dynamic input data
• Systems supporting such dynamic environments
• Performance Engineering
• Application development and run-time support

9
Past Practices for Applications and Systems
Design
Manufacturing Waterfall Model
Computer world Throw over the fence
10
Future Practices for Applications and Systems
Design
11
Past Design Methodologyfor computing systems

• Mostly ad hoc solutions many bright solutions
• Modeling and simulations for individual
  components
• Isolated islands of methods and tools
• No notion of modeling system performance
• no path from component to system behavior
• no path from analysis to prediction
• Performance measurements
• for workload characterization
• to detect bottlenecks/bugs
• mostly an afterthought
• Systems too complex to follow the same
  practices today

12
Systems Software/Hardware Architectural Framework
Applications/Users
Application
Collaboration Environments
Visualization
Languages
API Runtime Services
Compilers
Scalable I/O Data Management Archiving/Retrieval
Services
Authentication/ Authorization Dependability Servi
ces
Libraries
Tools
Other Services . . .
Global Management
Operating System
Distributed, Heterogeneous, Dynamic Computing
Platforms and Networks
Computing Engine
Device Technology
. . .
CPU Technology
Memory Technology
Components Technology
13
Distributed Systems Software/Hardware
Architectural Framework
Distributed Applications
Collaboration
Visualization
Environments
Authenication
/
Scalable I/O
Data Management
Authorization
Archiving/Retrieval
Dependability
Performance Engineered Design
Technology
Services
Services
. . .
Other Services
Distributed Systems Management
Distributed, Heterogeneous, Dynamic, Adaptive
Computing Platforms and Networks
Components Technology
Device
CPU
Memory
. . .
Technology
Technology
Technology
14
Distributed Systems Software/Hardware
Architectural Framework
Components Technology
15
Multiple views of the system The applications
view
Distributed Applications
. . .
Collaboration
Visualization
Environments
Authenication
/
Scalable I/O
Data Management
Authorization
IO / File
Archiving/Retrieval
Dependability
Models
Services
Services
. . .
Other Services
OS
Scheduler
Distributed Systems Management
Models
Architecture /
Distributed, Heterogeneous, Dynamic, Adaptive
Network
Computing Platforms and Networks
Models
Memory
Device
CPU
Memory
. . .
Models
Technology
Technology
Technology
16
Multiple views of the system The Operating
Systems view
Distributed Applications
. . .
Collaboration
Visualization
Environments
Authenication
/
Scalable I/O
Data Management
Authorization
Archiving/Retrieval
Dependability
Services
Services
. . .
Other Services
Distributed Systems Management
17
Performance Engineering

• Methodology
• consider the system in terms of its architectural
  layers
• multilevel, multi-resolution and multi-modal
  approaches
• combine different methods of describing
  components and layers
• ability to describe the system in multiple levels
  of detail (characteristics and time-scales)
• Performance Frameworks
• combine tools in plug-and-play fashion
• multiple views of the system

18
Performance Technology Innovations Needed

• Modeling languages for modeling and specification
  of performance attributes for such components and
  layers
• applications, system software, hardware
• Methods of modeling and simulation at multiple
  levels of detail and abstraction,
  models/simulators embodying these methods
• Ability to combine such multilevel and multimodal
  models and simulation tools (different levels of
  detail, different time-scales)
• Interfaces so for combining into the performance
  frameworks models and simulators of different
  resolution levels
• Measurement methods and tools
• instrumentation methods to address heterogeneity
  (systems and time-scales)
• dynamic integration with performance analysis and
  prediction
• storage and retrieval of measurements and other
  performance data
• Integration of these technologies into
  performance frameworks

19
(No Transcript)
20
Role of Performance Technology for System
Management

• Impact of lack of performance engineering tools
• managing distributed networks, systems,
  databases and applications remains a chaotic
  jumble of uncoordinated activities
• (the company) IT will spend half is budget on a
  SWAT team to keep systems running
• (the company) IT will spend an additional
  quarter of its budget in performance planning

Forrester Research Report, February
1998 Interviews with IT managers of 40 Fortune
1000 companies
21
Integrated Environments for Application
Development and Run-time Support (A Component
of the NGS Program)
22
Distributed Computing Support
Empowering Applications to exploit Future
Distributed Heterogeneous Computing Systems
Enterprise/Scientific/Engineering Computing
Impact
DisCoS

• Distributed compiler and programming model
• Application assembly technology
• Computing system analysis technology

Management enhanced with distributed applications
Example Target Recognition

• Distributed execution
• Highly efficient execution
• Real time or faster than real time
• Improve accuracy of analysis

• Enabling applications to efficiently execute
  across
• distributed, heterogeneous, and
  petaflops platforms

Large, Complex, Heterogeneous Applications
Defense Applications
Improved process for application design, support
and upgrade
Distributed Computing Support (DiSCoS)
Other Runtime Services
Example Pattern Recognition

• Reduce time to port applications
• from months to hour
• Reduce cost to port applications
• from 1M/port to 1K
• Enable rapid prototyping
• Reduce prototyping costs

Distributed Systems Management
Complex Distributed Adaptive Platforms
Distributed Computing Platforms and Networks
23
Technology Gap Example case Distributed
Application
Dynamic Analysis Situation

• Network of
• Workstations
• (NOW)

• Symmetric
• Multiprocessor
• (SMP)

Launch Application(s)

• Application cannot
• be repartitioned dynamically
• when problem size or
• number of SMPs changes

• Cluster of
• SMPs

Distributed Computing Resources
Adaptable Systems Infrastructure
24
DisCoS Technology for an integrated feedback and
control compiling system
DisCoS
Application Model

Dynamic Analysis Situation
Distributed Programming Model
Application Program
Compiler Front-End
Application Intermediate Representation
Compiler Back-End
Launch Application (s)
Architecture Models
Dynamically Link Execute
Application Components
Distributed Computing Resources
Distributed Platform
Adaptable computing Systems Infrastructure
25
Systems Software/Hardware Architectural Framework
Applications/Users
Application
Collaboration Environments
Visualization
Languages
API Runtime Services
Compilers
Scalable I/O Data Management Archiving/Retrieval
Services
Authentication/ Authorization Dependability Servi
ces
Libraries
Tools
Other Services . . .
Global Management
Operating System
Distributed, Heterogeneous, Dynamic Computing
Platforms and Networks
Computing Engine
Device Technology
. . .
CPU Technology
Memory Technology
Components Technology
26
Distributed Systems Software/Hardware
Architectural Framework
Distributed Applications
Application
Distributed Compiler and Run-time Support
Collaboration Environments
Mission Support Environments
API Runtime Services
Shared Storage Models
Security Dependability Survivability Services
Other Services . . .
Global Management
Distributed Systems Management
Distributed, Heterogeneous, Dynamic,
Adaptive Computing Platforms and Networks
Computing Engine
Device Technology
. . .
CPU Technology
Memory Technology
Components Technology
27
Application Composition System Challenges and
Approaches

• Present approaches for application software reuse
  and composition
• Libraries of application kernels
• Libraries for specific models of memory hierarchy
  
• Problem Solving Environments enable wiring
  together specific application
• Problems not addressed by existing technology
• Find and select compatible software components to
  build applications for heterogeneous platforms
• Bridge different data models used by components
• Build applications by dynamically composing
  independently-developed components

• New technology
• user interface libraries
  knowledge-base

• Straightforward extensions of existing
  technology
• Develop knowledge-based systems of components for
  specific defense applications
• Populate the knowledge data-base of components
  for specific platforms
• Mid-???
• Develop efficient data exchange mechanisms
  between different data representations
• Use data-mining to extract performance knowledge
  of specific application components
• Develop general interface mechanisms for
  selecting suitable components
• High Risk
• Automatic generation of application beginning
  from high-level specifications (e.g. text,
  equations)

28
Technology Areas (Supported by the NGS Program)

• Application Programming System (APS)
• distributed programming models and compilers
• Application Composition System (ACS)
• dynamic selection of distributed application
  components
• Application Analysis System (AAS)
• technology for performance engineered distributed
  applications
• Validation, Integration and Demonstration
• validation, integration and demonstration of the
  technology

29
Technology Roadmap
Application Programming System


Distributed programming models
.

Application performance Interfaces
.
.
i

Compilers optimizing mappings on complex
systems
n
t
D
Providing

Application Composition System
E
E
enhanced
g

Automatic selection of solution methods
.
.

Interfaces, data representation exchange
M
capabilities
.

Debugging tools
O
r
for
S
applications
a
t
Application Analysis System

i
o
.

Application/system multi-resolution models
.

Modeling languages
.

Measurement and instrumentation
n
Y1
Y2
Y3
Y4 Y5
Exploratory
Development
Integration Demos
Downselect
30
Agency Efforts

• NSF
• NGS The Next Generation Software Program
• SES Scalable Enterprise Systems
• ITR Information Technology Research
• Broad scope (take advantage of this!)
• Proposal for new initiative on
• Symbiotic, integrated simulations and
  measurements
• leap-ahead initiative
• will provide a focus for new exciting work in
  applications areas and
• in systems areas
• Also DARPA and NASA

31
Dynamic, Integrated Simulations and Measurements

• proposal to create new program
• potential to create leap-ahead technologies
• will create new paradigm for simulations
• will create new kinds of applications
• will provide a focus for new exciting work in
  applications areas and in systems areas
• new application algorithms
• active middleware services
• could involve multiple Directorates at NSF and
  perhaps others government agencies
• could involve industry

32
What about Industry

• Desktop is the driver for commercial software,
  and industry focuses on producing flexible
  software for the low-end
• Commercial/Enterprise Computing also poses
  requirements for more flexible and adaptable,
  reconfigurable interoperating systems and
  applications
• ..... BUT.....
• Industry focused on the short term returns,
  rather than investing on research for the
  enabling technology
• Moreover, industry addresses the problem by
  providing services for
• application porting
• application integration
• .... and making LOTS of s!!!
• Services is the fastest growing component of
  infotech business

on the other hand...
33
What about Industry (contd)and the Role of
Federal Agencies

• Industry however has history of adapting and
  productizing research and technology which has
  demonstrated success
• Need to strengthen the joint academe/industry
  research collaborations
• joint projects
• Technology transfer
• establish path for tech transfer from academic
  research to industry
• joint projects, students, sabbaticals (academe
  lt----gt industry)
• Initiatives from the Federal Agencies
• Effort analogous to one that pushed the frontiers
  for VLSI, Networking, and Parallel and Scalable
  computing
• Cross-agency co-ordination