Given by Geoffrey C. Fox at MPPOI 95 Conference on 24 October 95. Foils prepared 10 November 1995
Abstract * Foil Index for this file
See also color IMAGE
This talk stresses that main opportunity for optical interconnects in HPCC is in MIMD systems varying from the full National Information Infrastructure to a tightly coupled (petaflop) MPP |
We describe the WebWork project at NPAC, Cooperating Systems and Boston University which is studying use of Web Software Technology to integrate these two limits |
This table of Contents Abstract
Invited Presentation |
October 23-24 1995 |
Menger Hotel |
San Antonio Texas |
Geoffrey Fox |
NPAC |
Syracuse University |
111 College Place |
Syracuse NY 13244-4100 |
This talk stresses that main opportunity for optical interconnects in HPCC is in MIMD systems varying from the full National Information Infrastructure to a tightly coupled (petaflop) MPP |
We describe the WebWork project at NPAC, Cooperating Systems and Boston University which is studying use of Web Software Technology to integrate these two limits |
There are appear to be two natural ways to use optical connections for HPCC |
Firstly to support the National Information Infrastructure -- a closely connected set of computers |
Secondly to implement communication on a closely coupled MPP |
These are the national metacomputer versus local computer versions of "same" concept |
In this presentation we point out that regarding software as a commodity just as hardware suggests that we should build software for NII and closely coupled MPP's from same basic commodity parts --
|
The Petaflop conferences indicated that optical interconnections could well be essential to reach this performance even though clearly current (1995 Intel) Teraflop plans do not need optical connections for MPP supercomputers |
Note this analysis does not say key problem is SIMD Image processing using optical computing but rather supporting general MIMD systems with optics |
There are some interesting architectural issues in optical computing when you study NII as a "computer" as it has modest "bisection" bandwidth indicating data locality will be critical.
|
See Chapter 6 of Petaflops Report -- July 94 |
See Chapter 6 of Petaflops Report -- July 94 |
This architecture generalizes cutrrent IBM SP-2 type system and requires unlike Category I, data locality for the upto 40,000 CPU's to be able function efficiently with minimum communication overhead |
See Chapter 6 of Petaflops Report -- July 94 |
This design is an extrapolation of systems such as the J machine(Dally), Execube (Loral) or Mosaic(Seitz). It features CPU and memory integrated on the chip (PIM). |
Unlike such systems today, in the year 2015 such PIM designs have substantial memory per processor |
No silver programming bullet -- I doubt if new language will revolutionize parallel programmimng and make much easier
|
Social forces are tending to hinder adoption of parallel computing as most applications are areas where large scale computing already common
|
ATM ISDN Wireless Satellite advancing rapidly in commercial arena which is adopting research rapidly |
Social forces (deregulation in the U.S.A.) are tending to accelerate adoption of digital communication technologies
|
Not clear how to make money on Web(Internet) but growing interest/acceptance by general public
|
Integration of Communities and Opportunities
|
Technology Opportunities in Integration of High Performance Computing and Communication Systems
|
New Business opportunities linking Enterprise Information Systems to Community networks to current cable/network TV journalism |
New educational needs at interface of computer science and communications/information applications |
Major implications for education -- the Virtual University |
This is both Grand Challenges augmented by National Challenges but also |
Build HPCC technologies on a broad not niche base starting at bottom not top of computing pyramid |
1:Computational Fluid Dynamics |
2:Structural Dynamics |
3:Electromagnetic Simulation |
4:Scheduling |
5:Environmental Modelling (with PDE's) |
6:Environmental Phenomenology |
7:Basic Chemistry |
8:Molecular Dynamics |
9:Economic Modelling |
10:Network Simulations |
11:Particle Transport Problems |
12: Graphics |
13:Integrated Complex Systems Simulations |
14:Seismic and Environmental Data Analysis |
15:Image Processing |
16:Statistical Analysis |
17:Healthcare Fraud |
18:Market Segmentation |
Growing Area of Importance and reasonable near term MPP opportunity in decision support combined with parallel (relational) databases |
19:Transaction Processing |
20:Collaboration Support |
21:Text on Demand |
22:Video on Demand |
23:Imagery on Demand |
24:Simulation on Demand (education,financial modelling etc.) -- simulation is a "media"! |
MPP's as High Performance Multimedia (database) servers -- WebServers |
Excellent Medium term Opportunity for MPP enabled by National Information Infrastructure |
25:Military and Civilian Command and Control(Crisis Management) |
26:Decision Support for Society (Community Servers) |
27:Business Decision Support |
28:Public Administration and Political Decision(Judgement) Support |
29:Real-Time Control Systems |
30:Electronic Banking |
31:Electronic Shopping |
32:(Agile) Manufacturing including Multidisciplinary Design/Concurrent Engineering |
33:Education at K-12, University and Continuing levels |
Largest Application of any Computer and Dominant HPCC Opportunity |
1992: Grand Challenges |
1993: Grand Challenges |
1994: Toward a National Information Infrastructure |
1995: Technology for the National Information Infrastructure |
1996: Foundation for America's Information Future |
Applied Fluid Dynamics |
Meso- to Macro-Scale Environmental Modeling |
Ecosystem Simulations |
Biomedical Imaging and Biomechanics |
Molecular Biology |
Molecular design and Process Optimization |
Cognition |
Fundamental Computational sciences |
Grand-Challenge-Scale Applications |
Digital Libraries |
Public Access to Government Information |
Electronic Commerce |
Civil Infrastructure |
Education and Lifelong Learning |
Energy Management |
Environmental Monitoring |
Health Care |
Maunfacturing Processes and Products |
Application Specific NII Specific Services for
|
Like UNIX or MS-DOS or Windows 3.1(NT,95), WebWindows is an operating system for a "computer" |
The "computer" is a metacomputer consisting of the 50,000 Webservers (currently--eventually hundreds of millions) on Internet for the World Wide Web |
WebWindows can also be used for the metacomputer (collection of heterogeneous networked computers) which is a business enterprise system
|
WebWindows is a multi-client multi-server technology
|
It does not provide multi-threading/multiu-user support, memory management, device drivers and such base services -- these are supplied by UNIX, Windows or Mac O/S |
Rather it provides equivalent of higher level O/S services such as available under UNIX shell or applications supplied under Windows |
In the future one will build applications for WebWindows not UNIX / PC windows etc. |
Very interesting is WebWindows version of Lotus Notes to support Business Enterprise systems -- build from Web components such as those prototyped in WebTools
|
WebTools -- Early NPAC Prototype of WebWindows Equivalent to Program Manager with Navigation, File manipulation, Mail |
WebDeskTop Publishing -- an early killer application under WebWindows supplanting Word, Wordperfect, LOTUS123 , Persuasion etc. Java allows clear powerful implementation. |
WebRDBMS -- Integration of Relational and Distributed databases with both agent based heuristics, formal indices and free text search |
Metadata -- Common attributes to allow integration and search of heterogeneous databases |
Oracle 7 Interface to Usenet-Prepared October 27,1995 |
Associated material may be found starting at Oracle-Web Interface to Usenet and other Services |
Produced by Gang Cheng April 1995 |
There is a larger Better Quality Image available |
Produced by Gang Cheng April 1995 |
There is a larger Better Quality Image available |
Desktop Publishing and Productivity Tools in WebWindows (WebTop Productivity/Publishing)
|
InfoVision: Delivery on Demand of Information from:
|
Commerce -- digital Cash and signatures with appropriate authentication and security. Enables both Web Commerce (shopping) and use of Web for proprietary information |
Each of three components (network connections, clients, servers) has capital value of order $10 to $100 Billion |
InfoVision is ultimate "client-server" application
|
Democracy on the NII (Gore)
|
Web Servers use "Web Technology" to service World Wide Web and other forms of networked multimedia information |
All the News and Sports Archives of Reuters correspond to about 25,000 hours of information
|
Estimate similar storage needs for:
|
Total storage about 100 terabytes today |
This is basis of Virtual University and NII can be expected to be much more succesful than videosystems because interactive, collaborative, explorative and full multimedia information at student and teachers fingertips |
Base of Distance Education is InfoVision for full interactive curricula
|
Rich Collaboration environments including televirtual MOO's for student-student and student-teacher interactions |
Searched on Inventions |
Using San Diego VRML Viewer Webview |
Little Neck Bay in Northern Long Island (altitude exaggerated by factor 7) |
From Living Schoolbook Project |
Hot buttons linking to weather page in Albany area |
From Living Schoolbook Project |
A NYNEX Joint Venture |
This shows fiber draping Africa with coast off ramps |
Interactive and "batch" Collaboration
|
Metacomputing -- the collection of world wide computers as a coordinated (in subgroups of computers) computational engine (for simulation or information processing)
|
Critical feature is closely integrated collaboration of very many people and computers requiring |
Metacomputing Simulation and distributed database support as in Collaboratory |
Workflow support including configuration management and central CAD databases |
Rich collaboration environment to support distributed design decisions |
Standards and Security to allow interlinking of people and software from different organizations |
Key special requirement is distributed coordination |
A set of manufacturing companies -- Rockwell International, Northrop Grumman, McDonnell Douglas, General Electric and General Motors is studying the NII implications for a particular MAD system "Affordable Systems Optimization Process" (ASOP) |
Interesting parameters are that next major aircraft to be built could involve:
|
A sample VRML page produced by Black Hole Simulation group at NCSA and displayed by SGI WebSpace, cooperating with Netscape Navigator. |
A set of 3D spaces related to gravity research is represented as Netscape icons and linked to the corresponding VRML worlds. |
WebSpace window displays one of these spaces -- a space-time diagram for two black hole collision ('Pair of Pants' diagram) |
WebWork is an open, world-wide distributed computing environment based on computationally extended Web Technologies |
The backend computation and information infrastructure is provided by the World-Wide Virtual Machine -- a mesh of computationally extended Web Servers (called Compute Servers) |
These servers manage (via CGI mechanisms) a collection of standardized computational units called WebWork Modules. |
Geographically distributed and Web-published WebWork modules interact by HTTP/MIME based message/object passing and form distributed computing surfaces called Compute-Webs |
The front-end user/client interfaces are provided by evolving Web browsers with increasing support for two-way interactivity (e.g. Java, VRML) that facilitates client side control and authoring. |
A natural user-level metaphor -- WebFlow -- is supported in terms of visual interactive compute-web authoring tools. |
Implements the "Viable Base" Enterprise Model of HPCC Software identified in Pasadena2 workshop |
This will allow good programming tools to be developed and mnaintained as larger enough base to support software industry |
Implements a powerful software engineering framework for parallel computing by integrating parallel programming with the World Wide Web Productivity Tools |
WebWork is based on a three-layer architecture shown in figure 2, including: World_Wide Virtual Machine (WWVM) in the (bottom) layer 1, Middleware layer 2 of agents, wrappers, mediators etc., and high level programming environments (e.g. HPFCL) and user interfaces (e.g. WebFlow) in the (top) layer 3. |
All base WebWork concepts can be implemented in terms of today's Web technologies (HTTP, MIME, CGI) and a prototype is under development at NPAC. |
The overall design is open and ready to upgrade the existent (e.g. browsers or servers) and include new (e.g. agents or distributed object brokers) Internet/Web technologies |
One starting point for the WebWork construction is provided by NPAC WebTools -- a CGI-extended Web server with enhanced content authoring and database navigation functionalities. WebTools Server is used as a prototype WebWork node server. |
Illustrates 3 base layers of WebWork architecture and all main system components. |
A 4--node compute-web is represented
|
Java/HotJava model is used for WebFlow front-end implementation |
The paper describing this project is available at SCCS715 in NPAC technical report series |
Illustrates implementation of WebWork message passing in terms of
|
This diagram illustrates point-to-point communication between Web servers, used to implement a webflow channel between compute-web modules. Two extreme implementation modes are described: a) based on today's Web server technology, and b) based on thread memory mapped high performance implementation, expected in future Web compute-servers. Subsequent steps, represented by a sequence of labelled lines in the figure, are described below in both implementation modes. |
a) Today's Web server mode: (1) -- M1 locks O1 on S1 disk. (2) -- M1 sends POST HTTP message to S2 with M2 URL in the header sector and with O1 URL in the body sector. (3) -- S2 activates M2 via CGI and passes O1 URL as a command-line argument. (4) -- M2 sends GET method to S1 with O1 URL in the header. (5) -- S1 fetches O1 from its document tree. (6) -- S1 sends the content of O1 to M2 which completes the GET exchange. (7) -- M2 saves O1 by overwriting current I2 on the S2 disk. If I2 is locked, M2 waits (blocks). (8) -- After O1 is saved on the S2 disk, M2 returns 'end-of-transfer' acknowledgment to M1 which completes the POST exchange. (9) -- M1 unlocks O1 and exists. |
b) Compute-server (future Web server) mode: (1) - M1 locks its memory object O1. (2) - M1 checks if socket connection to M2 is in M1 connection table. If yes, go to (5) below. Otherwise, M1 connects to S2 and sends M2 creation script. (3) - S2 spawns M2 and acknowledges. (4) - M1 receives acknowledge message and saves new socket in connection table. (5) - M1 gets O1 handle. (6) - M1 writes O1 to M2 using socket lib calls. (7) - M2 reads O1 using socket lib calls. If I2 is free, O1 buffer is copied directly to I2 buffer. If I2 is locked, M2 creates O1 clone and blocks. (8) - M2 sends acknowledge to M1. (9) - M1 unlocks O1 and blocks. |
Latest results prepared for HPDC95 Tutorial August 1,1995 |
User-level WebWork metaphor is given by WebFlow -- a distributed dataflow model built in terms of WebWork modules and MIME object/document communication channels.
|
WebWork users will build and control distributed computing applications (compute-webs) using Web browsers based visual interactive editors and monitors. |
We are currently prototyping such WebFlow front-ends at NPAC using Java/HotJava model. WebWork modules are represented by Java threads (Screen 6) and visualized as interactive interconnected icons (Screen 7) |
Early prototype of AVS or Khoros like visual compute-web editor. |
Two interactive modes are supported:
|
In mode a), each click in the active editor window places a new module box there. |
In mode b), each click on module port generates links with all other modules. |
One current WebWork/WebFlow application, prototyped at NPAC, is Software Project Manager (Screen 8). Each software developer runs his/her WebTools server and uses HySource CASE tools. These servers are WWVM-connected to agent and manager servers. Agent server receives automatic notifications from developers servers on each software volume update, and uses customizable thresholds to decide when to fire a report to the manager or a deadline reminder to a developer. |
Software Project Manager tools contains a simple agent server that mediates between client/consumer ( here manager) and servers/producers (here developers). |
A front-end for the software project manager tool. |
Three types of modules are supported:
|
Developer modules are linked to the agent module and report automatically all changes in the software volume (handled by WebTools CASE toolkit integrated with WebTools editor). |
The agent module integrates the results and uses customizable threshold to decide when to fire a report to the manager or a deadline reminder to a developer. |
More generally, this Middleware Layer 2 will be rather complex and populated by a spectrum of proprietary (e.g. Telescript, ScriptX, CORBA) and public (e.g. Perl, Tcl, Harvest, Java, VRML) scripted languages, brokers, agents, wrappers, mediators etc. see Screens |
In WebWork, we refer collectively by WebScript to the whole ensable of these models. |
At the current stage, it isn't clear if WebScript as a common intermediate language is a practical concept. An alternative is to live in the multi-language Web medium and emply interoperability agents to translate between various protocols. |
Practical initial implementation platfrom for this dual approch is provided in WebWork by an integrated collection of WebTools CASE tools based HySource Worlds for various languages. |
HyPerl World page, generated automatically by the WebTools CASE package, and integrating documentation with the source. |
More generally, we call by HySource the hypertext documentation with navigable source code included. |
Function calls and external variable references are 'blue' and point to the corresponding HySource pages. |
RSA155 requires about 300 teraops hours to solve with NFS |
RSA129 needed about an order of magnitude less time. Can be done today faster if use Number Field Sieve |
We have roughly one to five million independent calculations which form the rows of matrix (after clever graph theory manipulates and combines) |
Set of master servers publish problem to solved with suitable demos, description of algorithm and full marketing attention.
|
Clients return results -- not so easy except by email and cut and paste |
Best done as a set of cooperating servers where server performing factorization publishs it solution as a file on the WWW. |
Cooperating servers also better for computer administrators as can control set of clients at a given site |
Initially use humans but replace by agents when software ready |
In WebWork, we also propose an interpreter of HPFCL -- High Performance Fortran Coordination Language, which will support coarse grain distributed HPF computation. Compiled HPF modules, published on individual nodes of the WWVM, will be easily invoked by HPFCL scripts, integrated with GUI front-ends (such as Khoros etc.) and employed in collective computation on the WWVM. |
In WebHPL we further explore the concept of interpreted HPCC language environments and we propose an object-oriented Web based parallel programming environment supporting HPF and C++ for distributed metacomputing. |
WebHPL, or Web based High Performance Languages, is our most ambitious project in the area of Web and HPCC integration. It addresses both base software engineering and applications, and it refers both to backend and frontend layers of language compilers and interpreters, seeking a uniform programming model for interactive HPCC. |
WebHPL is a hybrid Compiled/Interpreted environment with a more or less seamless use interface. |
We use compilers or optimized message passing in the local environments where low latency hardware benefits from careful optimization. |
We use interpreters where flexibility and power (e.g. fault tolerance) of agent approach beneficial and performance overhead in software matches intrinsic lower performance of hardware. |
We expect that our two-prong approach in WebHPL, attacking the problem both from the full compiler and the full interpreter perspective, will result in an interesting software engineering framework, 'scalable' along the compiler<--->interpreter axis. |
The interpolation mechanism along this axis is provided in terms of domain specific "little languages" that are interpreted but operate on compiled modules. UNIX shell, Perl or Tcl are examples of sequential little languages. WebHPL will develop parallel analogs. |