Some Implications of the Web for IBM

Technical Advisory Board Meeting Feb 6-7, 1996.

Jack Dongarra

Geoffrey Fox

Tony Hey

Mal Kalos

Cherri Pancake

Rick Stevens

Introduction

1996 sees continued rapid progress and change in the Web. However enough capabilities and technologies have been developed (possibly only in prototype form and with more or less formal agreement) that it is now reasonably clear how one can and should develop applications "on top of the Web". We anticipate a growing availability of software products in fields such as "WebTop" Productivity tools, finance, health- care, business enterprise support. We strongly recommend an aggressive program by IBM to provide appropriate systems support for this emerging area which could become the dominant computing paradigm in only a few years. Such paradigm changes only occur rarely and IBM cannot afford to wait before setting up its program in this area. In the following, we describe some features of the emerging Web which seem important in setting up IBM's strategy. Then we give a few recommendations.

Some Features of the Web Computing Revolution

(1) Some Comments on Current Status of Web and its Technology

The Web has evolved rapidly with many commercial, government and academic institutions exploring and enhancing its capability as a pervasive information hardware and software technology. Note that when we talk about "the Web", we do not imply the use of the Internet which is not secure and has modest performance. Rather we can use technologies such as Java, HTML, and VRML which have largely been developed for an Internet clientele, to build systems that use any network infrastructure. This can be mobile, satellite, ATM, ISDN, public, private or have any mixture of such characteristics. See a recent Business Week article on IntraNets for an interesting discussion and some projections . Information week has some similar predictions in growth of Web Server business with over 100% growth per year.

During 1995, many key Web technologies emerged: Java which can be expected to become the dominant distributed computing language; VRML 1.0 as a descriptor of 3D objects; integration of databases and the Web; Netscape Chat and other early simple collaborative tools; scripting languages such as Livewire (server side) and JavaScript (client side). Further we saw several examples of what we call "WebWindows" - namely application software that is written to the interfaces defined by Web server, client and data transport standards and which we discuss below.

(see IBM Tutorial on Web Technology for HPCC given february 7,96 by Geoffrey Fox for a general discussion)

(2)The WebWindows Concept

WebWindows will be the Operating System of the Future. The Web will allow the development of both new and better (old) applications. This will make a major change in the software industry as the Web naturally supports modular applications with generally many small companies developing individual components with "Web Integration" businesses supporting and developing the full application. Not only will monolithic (stovepipe) applications (such as Microsoft Word) disappear but the role and structure of operating systems will change. Traditional operating systems for individual systems will not need to support users directly but rather just Web Servers and Clients. Future software will be written not for UNIX, the Win32 interface to Windows 95/NT etc. but rather for "WebWindows". We expect the Web Windows approach - now illustrated by WebMail, WebDatabases, WebEditors, WebFoil (NPAC) and multimedia products such as Real Audio and VDOLive - to grow in popularity.

A system built under WebWindows is intrinsically plug and play - one develops the full system with modules - interfaced at the communal Web standard - taken from many different sources. It is critical to design one's system to use appropriate modules - in particular, one must build a modular and not a monolithic system. As one listens to the "drumbeat" of the Web (typically by monitoring key news groups), one will find and explore new modules which can be inserted into the WebWindows environment.

WebWindows is defined by the use of Web Servers and clients, their open interfaces (HTML, VRML, CGI etc.) and their technologies (Java, LiveMedia etc.). ISV's and users will judge hardware platforms by their support of the "WebWindows" virtual machine. For instance, there will be a set of "WebWord" products available which will offer the functionality (and more) of Microsoft Word (and similar products) but be built with Web technologies. WebWord is quite different (in architecture and hence implementation) from linking Word to the Web(Weberizing Word) and so Microsoft has no special advantage in the WebTop Productivity field. IBM itself has similar potential difficulties with Lotus Notes.

(3) Server-Server Model of Computing

The Web will not only allow new technologies to develop better versions of current applications (such as Word) but also produce new uses of computers. The current Web roughly implements a client-server computing model but as client-side web technologies improve one sees an evolution to a new paradigm - a server--server computing model. This does not say that clients will disappear but rather they will evolve to incorporate much of the functionality now in servers. In fact current clients are much larger software packages than most servers! Already the Web links some 50,000 servers together but technologies such as Java and a growing use of interactive and collaborative applications will either directly (run a server on client machine) or indirectly (enhanced browsers) turn all clients into servers. This implements the democratic Web where all of us can be consumers and producers of information for the Web. Remember that when we talk of products using Web technology, it does not have to be applied to the full world wide Internet. Rather WebWord could run a single PC which has both server and client capabilities. Again Web linked DB2 and a future WebNotes product could use Web Technology and IBM Webserver systems to support a Business Enterprise Information system on a closed corporate network. These are the IntraNets The Systems opportunity is support (perhaps with high availability capability) the "server infrastructure" which is a dynamic set of linked servers. This infrastructure would provide services such as collaboration support, information search, storage and dissemination, metacomputing, commerce etc.

(4)Structure of Web Software

Currently we can picture Web Software in following fashion

This picture comes with color from a foilset

In future, one might see a more modular:

Again this picture is available with color from a foilset.

In this last picture, we see a difficult challenge for producers of monolithic software system producers but an opportunity for IBM to produce optimized servers with Web Server support implemented in "native fashion" and not on top of traditional operating systems.

Some Opportunities for IBM

• 1)As a company, embrace the Web approach to computing. Participate actively in setting and evolving standards.
• 2)Accelerate development and deployment of high quality and high performance Java interpreters and Java runtime on single CPU RS6000, RS6000 SMP and SP distributed memory systems.
• 3)Similarly accelerate development and deployment of optimized servers, and VRML browsers on IBM platforms.
• 4)Develop a high quality multimedia Webserver offering for the full range of RS6000 to SP systems. This server should be consistent with Web standards such as LiveMedia for multimedia data and support well linkage of databases such as Oracle, DB2 and Informix(Illustra). IBM's High Availability (HA) technology could offer significant value here. Note the single (file) system image of SP offers significant advantages for a large video server over an loosely coupled set of Windows NT servers. Presumably current Yorktown work on video servers will be a good prototype to start from.
• 5)IBM has a excellent existing business customer base using mature mainframe technology. A dual strategy seems appropriate. Migrate these customers to SP based support of their current OLTP and other information needs. Grow their use of Web based information systems expecting this will eventually include OLTP, collaboration (WebNotes) and datamining support. Here datamining would include future versions of current Web Search activities.
• 6) The opportunity is not just a uniform Web Interface from RS6000 to SP2 but IBM can offer a much needed user transparency between SP2/RS000 and ES9000 and AS400 systems. For example, in a single (or adjacent boxes) you can have both the company critical TP application and the more speculative data-mining or OLAP or ... applications. IBM need to provide mechanisms for moving data, or at least making it accessible to both systems.
• 7) It does seem a good strategy for SP2 to emphasize its Web interface/OS and its reliability (compared to other Unix systems) via its Phoenix HA approach during the next few years until it regains floating point competitiveness. With the right software and marketing its reputation for reliability and forward looking Web compatibility, these features could outweigh the hardware deficiences.
• 8)There is an interesting opportunity to develop better parallel computing tools based on Web technologies. These would exploit linkage of information and computing which is natural on the Web. The high quality of Web applications should allow this approach to give better tools without increasing level of effort and so enhance the SP product line. This could be viewed as a cost issue -- it will be much cheaper to build systems with standard Web interfaces and internal Web technology . Not only will each component be able to build on pervasive technology but more importantly it will be very much easier to build systems as the linkage of components via Web interfaces will be built-in to the design without special attention in system architecture.
• 9)A major long term opportunity is the development of optimized hardware and software to support Web applications. The software opportunity is the integration of Web server services at a lower level to get better performance from a "native" implementation. The hardware opportunity is to examine needs of multimedia and other client and server services and build in hardware the appropriate systems support. Note this suggestion has an important implication for NT versus UNIX debate. It says that neither of these is future winning approach but rather one needs a fresh look at operating systems. It is believed that Sun intends to develop such native Web O/S using Java.