Full HTML for

Basic foilset A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG

Given by Geoffrey C. Fox, Yuhong Wen, Wojtek Furmanski, Tom Haupt at NASA Workshop on Performance Engineered Information Systems on Sept 28-29 98. Foils prepared Sept 30 98
Outside Index Summary of Material

PetaSIM Motivation and basic ideas
  • (Using aircraft analogy) aimed at conceptual design of computer architecture and applications ( as opposed to later preliminary and detailed design)
  • Java applet as friendly user interface; C++ execution engine
PetaSIM Design and Examples
  • as conceptual level, can use to estimate performance of applications in any language
  • Like RMI, use Java to specify object structure of computers and problems -- need to integrate with other PSL
Sample PetaSIM Experimental Results
A hybrid object web model for IPG (Information Power Grid)
Possible use of PetaSIM in IPG for two classes of applications
  • Multimedia transport in collaborative systems
  • Collection of coarse grain data flow modules

Table of Contents for full HTML of A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG

 Denote Foils where Image Critical
Denote Foils where HTML is sufficient
1 A Coarse Grain Performance Estimator PetaSIM and a hybrid distributed object model for IPG -- Information Power Grid NASA Performance Engineered Information Systems Workshop NASA Ames Mountain View September 28-29 1998
2 Summary of PetaSIM POW and IPG
3 PetaSIM Interpolates Between
4 Architecture of PetaSIM
5 Interface of PetaSIM Client
6 PetaSIM Design and Functionality
7 Petasim Estimator & Emulator
8 IBM SP2 Architecture I used in tests
9 IBM SP2 Architecture II used in tests
10 Nodeset: Sample Object Structure
11 Dataset: Sample Object Structure
12 Pathfinder Scaling: Performance v. # Processors
13 Pathfinder Estimation Results II
14 Titan Estimation Results (Fixed)
15 VMScope Estimation Results
16 Today's Pragmatic Object Web: The Confusing Multi-Technology Real World Middleware Server Layer
17 Summary of Pragmatic Object Web
18 Summary of NPAC's JWORB natural Building Block of the Middle Tier
19 IIOP Performance for Java ORB's
20 HPcc: High Performance commodity computing for the IPG: Information Power Grid
21 Hybrid Parallel Cluster Computing Model
22 WebFlow + Globus Functional Architecture
23 WebFlow over Globus for NCSA Alliance Quantum Chemistry Application View
24 HLA/RTI and Coarse Grain Technologies I
25 Metaproblem Hierarchy
26 HLA/RTI and Coarse Grain Technologies II
27 PetaSIM and the IPG I
28 PetaSIM and Collaborative Engineering I
29 PetaSIM and Collaborative Engineering II
30 Integration of PetaSIM with Collaborative Software
31 Data Flow Modules and PetaSIM I
32 Data Flow Modules and PetaSIM II
33 Some Relations to Other Projects

Outside Index Summary of Material

HTML version of Basic Foils prepared Sept 30 98

Foil 1 A Coarse Grain Performance Estimator PetaSIM and a hybrid distributed object model for IPG -- Information Power Grid NASA Performance Engineered Information Systems Workshop NASA Ames Mountain View September 28-29 1998

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Geoffrey Fox, Wojtek Furmanski, Tom Haupt and Yuhong Wen
Northeast Parallel Architecture Center (NPAC)
Syracuse University
gcf,furm,haupt,wen@npac.syr.edu
URL is http://kopernik.npac.syr.edu:4096/petasim/V1.0/PetaSIM.html
HTML version of Basic Foils prepared Sept 30 98

Foil 2 Summary of PetaSIM POW and IPG

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
PetaSIM Motivation and basic ideas
  • (Using aircraft analogy) aimed at conceptual design of computer architecture and applications ( as opposed to later preliminary and detailed design)
  • Java applet as friendly user interface; C++ execution engine
PetaSIM Design and Examples
  • as conceptual level, can use to estimate performance of applications in any language
  • Like RMI, use Java to specify object structure of computers and problems -- need to integrate with other PSL
Sample PetaSIM Experimental Results
A hybrid object web model for IPG (Information Power Grid)
Possible use of PetaSIM in IPG for two classes of applications
  • Multimedia transport in collaborative systems
  • Collection of coarse grain data flow modules
HTML version of Basic Foils prepared Sept 30 98

Foil 3 PetaSIM Interpolates Between

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
"Back of the envelope estimates" where you get good intuition for why the performance is what it is
and Precise Simulations
Communication Overhead in classical data parallel (edge over area) with NO memory hierarchy NO latency where dim is geometric dimension and tcomm and tfloat are
typical communication and computation times
HTML version of Basic Foils prepared Sept 30 98

Foil 4 Architecture of PetaSIM

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Imagine engineer using applet to change system parameters interactively and examine new network/machine designs quickly -- easier to change system parameters than generate new applications -- so need to accumulate benchmark set
HTML version of Basic Foils prepared Sept 30 98

Foil 5 Interface of PetaSIM Client

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Specify Nodes
Specify Links
Specify Datasets
Specify Execution
HTML version of Basic Foils prepared Sept 30 98

Foil 6 PetaSIM Design and Functionality

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
We define an object structure for computer (including network) and data
  • These object representations can also be used in dynamic tools in areas of scheduling (eg. Condor, Legion or Globus) and seamless interfaces (e.g.. UNICORE, WebSubmit, SWeb)
Architecture Description
  • nodeset & linkset
  • Special attention to support of hierarchical memory machines and architectures to support data intensive applications
Data Description
  • dataset & distribution (not stressed in current state of development)
Application Description -- needs further refinement
  • execution script which is generated either by hand or automatically from Maryland
HTML version of Basic Foils prepared Sept 30 98

Foil 7 Petasim Estimator & Emulator

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
PetaSIM
Performance Estimation
Nodeset
Linkset
Dataset
Distribution
UMD Emulators Automatic Script Generation
Execution Script
Hand Coded
Script
Applications
HTML version of Basic Foils prepared Sept 30 98

Foil 8 IBM SP2 Architecture I used in tests

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
View as a bunch of nodesets joined by a bunch of linksets Each component defined as "objects" which are valuable outside PetaSIM (in defining object
structure of computers)
Only One
member of
most nodesets
shown
HTML version of Basic Foils prepared Sept 30 98

Foil 9 IBM SP2 Architecture II used in tests

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Just one member
of each CPU level
nodeset
shown in this
more detailed
Architecture
for SP2
HTML version of Basic Foils prepared Sept 30 98

Foil 10 Nodeset: Sample Object Structure

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Name: one per nodeset object
type: choose from memory, cache, disk, CPU, pathway
number: number of members of this nodeset in the architecture
grainsize: size in bytes of each member of this nodeset (for memory, cache, disk)
bandwidth: maximum bandwidth allowed in any one member of this nodeset
floatspeed: CPU's float calculating speed
calculate(): method used by CPU nodeset to perform computation
cacherule: controls persistence of data in a memory or cache
portcount: number of ports on each member of nodeset
portname[]: ports connected to linkset
portlink[]: name of linkset connecting to this port
nodeset_member_list: list of nodeset members in this nodeset (for nodeset member identification)
HTML version of Basic Foils prepared Sept 30 98

Foil 11 Dataset: Sample Object Structure

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Name: one per dataset object
type: of dataset choose from grid1dim, grid2dim, grid3dim
bytesperunit: number of bytes in each unit
floatsperunit: update cost as a floating point arithmetic count
operationsperunit: operations in each unit
update(): method that updates given dataset which is contained in a CPU nodeset and a grainsize controlled by last memory nodeset visited
transmit(): method that calculates cost of transmission of dataset between memory levels either as communication or as movement up and down hierarchy
  • Methods can use the other parameters or be custom
HTML version of Basic Foils prepared Sept 30 98

Foil 12 Pathfinder Scaling: Performance v. # Processors

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Measured Execution Time
PetaSIM Running Time
Estimated Application Execution Time
HTML version of Basic Foils prepared Sept 30 98

Foil 13 Pathfinder Estimation Results II

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Measured Execution Time
PetaSIM Running Time
Estimated Application Execution Time
HTML version of Basic Foils prepared Sept 30 98

Foil 14 Titan Estimation Results (Fixed)

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
PetaSIM Running Time
Estimated Application Execution Time
Measured Execution Time
HTML version of Basic Foils prepared Sept 30 98

Foil 15 VMScope Estimation Results

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Measured Execution Time
PetaSIM Running Time
Estimated Application Execution Time
HTML version of Basic Foils prepared Sept 30 98

Foil 16 Today's Pragmatic Object Web: The Confusing Multi-Technology Real World Middleware Server Layer

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
W is Web Server
PD Parallel Database
DC Distributed Computer
PC Parallel Computer
O Object Broker
N Network Server
e.g. Netsolve, Ninf
T Collaboratory Server
Clients
Middle Layer (Server Tier)
Third Backend Tier
HTML version of Basic Foils prepared Sept 30 98

Foil 17 Summary of Pragmatic Object Web

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
3-(or more)-tier architecture - Web browser front-ends, legacy (e.g. databases, HPC modules) backends; fat middleware
Use as appropriate the alternative / competing Middleware models:
  • Java RMI+ EJB (Enterprise Javabean) - single language solution by Sun
  • CORBA - all languages solution by OMG
  • COM - multi-language solution by Microsoft
  • WOM/XML - emergent solution by the Web Consortium
Each model has different tradeoffs (most elegant, powerful, fastest, simplest)
POW integrates various models and services either by linking multiple brokers/servers or in terms of a single multi-protocol middleware server (JWORB)
  • Note Java is often the best language to build middleware whether this is Java or some other distributed object model
HTML version of Basic Foils prepared Sept 30 98

Foil 18 Summary of NPAC's JWORB natural Building Block of the Middle Tier

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
JWORB - Java Web Object Request Broker - multi-protocol middleware network server (HTTP + IIOP + DCE RPC + RMI transport)
Current prototype integrates HTTP and IIOP i.e. acts as Web Server and CORBA Broker
  • HTTP Services built in terms of CORBA services
  • Gives you immediately web interfaces to CORBA
  • CORBA supports applications in any language
Next step: add DCE RPC support to include Microsoft COM
JWORB - our trial implementation of Pragmatic Object Web
First non DMSO implementation of RTI -- HLA (distributed event driven simulation) Runtime at 5% cost(!)
HTML version of Basic Foils prepared Sept 30 98

Foil 19 IIOP Performance for Java ORB's

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
JacORB
JWORB
ORBIX
RMI
Variable Size Integer Arrays
HTML version of Basic Foils prepared Sept 30 98

Foil 20 HPcc: High Performance commodity computing for the IPG: Information Power Grid

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Adopt multi-tier enterprise computing model
  • Client --> Middle Tier Server ---> Backend high performance system
Use Commodity Software Model as basic distributed computing infrastructure in middle tier
Use Pragmatic Object Web(POW) as the abstraction of today's commodity distributed object software infrastructure
Middle tier uses high functionality protocols such as Web HTTP, CORBA IIOP, Java RMI etc. with MPI (Nexus or Globus) as optimized "machine code" at backend
So we have a natural hierarchical model with perhaps an SP2 seen as a single POW distributed object in middleware and 128 nodes at backend
Nodes in a Cluster host JWORB server when seen at middle tier and this controls the MPI process view at backend
Note logical progression below which is opposite to some other approaches
HTML version of Basic Foils prepared Sept 30 98

Foil 21 Hybrid Parallel Cluster Computing Model

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Backend Parallel Computing Nodes running Classic HPCC -- MPI becomes Globus to use distributed resources
Middle Control Tier -- JWORB runs on all Nodes
SPMD Program
SPMD Program
SPMD Program
SPMD Program
MPI
JWORB
JWORB
JWORB
JWORB
RTI
Use separation of control and data transfer
to support RTI(IIOP) on control layer and MPI
on fast transport layer simultaneously
RTI
RTI
MPI
MPI
Middle and Backend on Each Node
HTML version of Basic Foils prepared Sept 30 98

Foil 22 WebFlow + Globus Functional Architecture

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Client Tier
Middle
Tier
IIOP
Backend: Globus MPI etc.
IIOP
HTML version of Basic Foils prepared Sept 30 98

Foil 23 WebFlow over Globus for NCSA Alliance Quantum Chemistry Application View

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
HTML version of Basic Foils prepared Sept 30 98

Foil 24 HLA/RTI and Coarse Grain Technologies I

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Consider large class of problems that can be thought of a set of coarse grain entities which could be internally data parallel and the coarse grain structure is "functional or task" parallelism
Use (Enterprise) JavaBeans to represent modules at (server) client level
Use UML (and related technologies) to specify application and system structure
WebFlow is graphical (Java Applet) composition palette (Beanbox for computational modules)
Use "To be Agreed Seamless Computing Interface" to implement linkage of proxies to backend hardware
We can support any given paradigm at either high functionality (web server) or high performance (backend) level
HPCC Messaging could be a pure Java/RMI middle tier version of MPI or Nexus/Optimized Machine specific MPI at backend
HTML version of Basic Foils prepared Sept 30 98

Foil 25 Metaproblem Hierarchy

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Full Heterogeneous MetaProblem
Module
Aggregates of grid points etc.
Module
Module
Module
Module
Components
Components
Task Parallelism
Data Parallelism in Loosely
Synchronized Computation
Split into
Levels
Memory Hierarchy
including I/O
Fine Grain Simulations
One or more
levels of
Coarse Grain
Simulation
PetaSIM can be used at either coarse or fine grain
Working at coarse grain, can abstract fine grain in a simple model or model in detail
HTML version of Basic Foils prepared Sept 30 98

Foil 26 HLA/RTI and Coarse Grain Technologies II

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Coarse Grain Entities can be time synchronized simulations and use MPI(HPF?) at either middle or back end tier or as in DMSO simulations a federate running a custom discrete event simulation
Use DMSO Object model HLA to specify object structure of jobs and systems at middle tier level
A HLA Federation could be the set of all jobs to be run on a particular site
  • We can classify both jobs and computers as separate federations
A HLA Federate could be a job consisting of multiple possibly shared objects
Use DMSO Runtime Infrastructure RTI to implement dynamic management
  • As RTI already incorporates time management as a service, it can be basis of powerful commodity resource management at level of linked coarse grain objects
HTML version of Basic Foils prepared Sept 30 98

Foil 27 PetaSIM and the IPG I

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
We can exploit hierarchical view of metaproblem and use PetaSIM to model the collection of "middle-tier" modules
  • Requires extension of linkset object to handle geographically extensive networks
  • Requires models of performance of modules which could be parallel simulations
Rather than support general case, consider two examples
  • Multimedia datastreams as in collaborative sessions
  • Dataflow linkage of a few computational / visualization / database modules
  • Their integration in "full" collaborative engineering
HTML version of Basic Foils prepared Sept 30 98

Foil 28 PetaSIM and Collaborative Engineering I

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
There are (at least) two types of data streams in collaborative engineering
  • Multimedia data -- typically audio - video streams from participants
  • Simulation results, Displays from planning tools (CATIA), Information from web pages
IPG: Real Time Multimedia and Asynchronous Data
HTML version of Basic Foils prepared Sept 30 98

Foil 29 PetaSIM and Collaborative Engineering II

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Collaborative Sessions have new measures of performance
Quality of received multimedia -- especially audio -- which is streamed in real time and data thrown out if comes too late
Results of simulations and resources (e.g. web pages accessed) used in session can be transmitted asynchronously. If network congestion, the results are typically correct but received late
  • We are familiar with and tolerant of this with the web
  • However in a collaborative session, if data delayed to one participant all have to wait or that participant does not have this information during discussion. One uses mirror sites and proxy servers to alleviate this problem
All of these features and implied goodness measures should be built into PetaSIM -- this appears straightforward
HTML version of Basic Foils prepared Sept 30 98

Foil 30 Integration of PetaSIM with Collaborative Software

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
This use of PetaSIM can be used upfront to deploy appropriate network for collaborative engineering
Further it can be used dynamically by collaboration tool to make decisions on
  • resolution to use for web images or video streams; codec and buffer size to use for audio
Collaborative software (such as TangoInteractive) can support API allowing monitoring of delays so PetaSIM can determine when material arrives at each engineers desk and adjust algorithm
HTML version of Basic Foils prepared Sept 30 98

Foil 31 Data Flow Modules and PetaSIM I

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
In example below, need to model performance of 6 modules
Visualization
HTML version of Basic Foils prepared Sept 30 98

Foil 32 Data Flow Modules and PetaSIM II

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
One or more Client Server Applications are a special case of Dataflow
  • Application servers in industry are moving client server to dataflow
Can be applied either to a geographically distributed network or to a cluster of PC's or workstations
In PetaSIM, need to define "middle-tier" objects and their performance parameters and "simple estimates" as in PetaSIM fits the coarse graining of applications
General problem is multiple "dataflow graphs" and collaborative sessions to be modeled in context of a given network model
  • MSTAR (as in Darpa Performance Program) is Dataflow (Khoros)
HTML version of Basic Foils prepared Sept 30 98

Foil 33 Some Relations to Other Projects

From A Coarse Grain Performance Estimator -- PetaSIM and a Hybrid Distributed Object Model for the IPG NASA Workshop on Performance Engineered Information Systems -- Sept 28-29 98. *
Full HTML Index
Ninf (and NetSolve) are examples of POW Middle tier servers
Coarse Grain Dataflow is an example of "general task graph" described in POEMS
Coarse Grain Dataflow and Collaborative Session are examples of AppLeS Templates
PetaSIM could be thought of as modeling tool for AppLeS (aimed at coarse grain quick runtime estimations)
Web and Object Servers used synchronously as in collaborative sessions have very high re-use and very different trade-offs (proxies, mirror sites more important) from classic asynchronous access
© Northeast Parallel Architectures Center, Syracuse University, npac@npac.syr.edu

If you have any comments about this server, send e-mail to webmaster@npac.syr.edu.

Page produced by wwwfoil on Sat Nov 28 1998