In-Depth

The Internet’s Impact on the Role of the Mainframe

• 09/01/1998

Once the mainframe was synonymous with computing. Then the enterprise computing model began to shift to client-server. In the heyday of client-server in 1991, pundits predicted the extinction of the mainframe within five years. But more mainframe MIPS were sold in 1996 than ever before in history and the growth rate is expected to continue increasing. And most analysts agree that over 70 percent of all corporate data still resides on the mainframe.

Today, a third enterprise computing model has combined features of the mainframe-centric model with the client-server model and IT is recentralizing again with the universal client model — where a Web browser provides the front-end to a business application and the business logic resides elsewhere (Source: Meta Group, 1996).

Corporate IT is now being challenged with integrating the Web and the mainframe, while leveraging their existing systems.

Multi-tier distributed computing is often seen as the as best way to migrate business logic off the fat clients of the client/server model and a host of buzzwords has evolved around the new middleware layer. But drawbacks to this scheme include the need for new infrastructure, the difficulty in providing scaleable applications and the greatly added complexity of the transaction routing.

Another approach to the distributed architecture of the universal client is two-tiered. In this implementation, the business logic remains on the host, just like in the mainframe-centric model. This provides an immediate solution that leverages the robustness and power of the mainframe. Many feel this to be a perfect blending of the two previous models — combining the scalability, reliability and security of the mainframe on the back-end with the flexibility of the PC on the front-end.

This approach is in line with the recent trend of host recentralization seen at many medium-sized to large companies. Host recentralization improves system performance, simplifies administration and reduces support costs. Recentralization also gives IT organizations a breather, while they wrestle with the complexity of n-tier distributed processing. Mainframes continue to be a robust and secure platform on which to run mission-critical business applications.

Using a Web browser in a two-tier solution requires that the Web services be hosted on the mainframe. Since MVS/CICS is the most prevalent production platform on the mainframe — the one platform with access to most mainframe applications and data — Web services directly within CICS provide significant advantages, such as the ability to leverage the investment in existing applications.

How do these three models — mainframe-centric, client/server and universal client — apply to corporate IT’s future planning? Much depends on the current status of each organization — where they are in the adoption cycle. End users could have 3270s connected to CICS applications over SNA; they may have PCs connected to a LAN using 3270 emulators to access CICS applications or they could have PCs running a "fat client" accessing data stored on the host. A number of companies have a mixture of all three environments within their corporate IT departments. There may well be a departmental need for each level of adoption within a large corporation.

Many companies with thousands of 3270 terminals now find they have a compelling business requirement to enable those end users to browse the Web — either to access the company intranet to review policies and procedures or even the Internet to gain access to timely information available nowhere else. Replacing those terminals would require a massive capital outlay to acquire PCs and to build a whole new infrastructure — not to mention all the new maintenance and training issues.

To best address this type of need, software is required that can be installed on the mainframe and enable each 3270 to function as a text-based Web browser. Since the vast majority of 3270 users are accessing CICS systems, it is logical that a 3270 Web browser also run under CICS. Mainframe TCP/IP and CICS sockets are becoming increasingly more common and is the only requirement for these CICS users to browse the Web.

Organizations that have replaced their 3270 terminals with PCs and emulators have a different set of needs. Each end user’s PC can browse the Web if provided with a TCP/IP stack, so accessing the intranet or Internet is less of an issue. Their legacy applications, however, still remain separate and apart from the Web.

The need in this case is for a new technology that can exploit the Web, while incorporating these legacy systems. One interesting technology allows Web-based graphics or multimedia to be delivered over the legacy CICS connection, interactive with the legacy systems.

This type of implementation can dramatically enhance the capabilities of a legacy system without having to modify any existing code. End users can instantly view the graphical image of a document associated with their underlying legacy screen. Uses include the display of an invoice, while running an accounts receivable application, the display of a detailed schematic from within a manufacturing or maintenance application or the display of a related document stored on an imaging system.

Ultimately, many organizations may wish to go even further and replace the 3270 interface altogether with the universal client. This type of implementation brings up several new issues.

There are several approaches to adapting legacy applications to run with a browser front-end. Dedicated screen scrapers and data replication approaches share the same scalability and security problems as does any multi-tier configuration. A mainframe Web server uses a two-tier architecture to eliminate scalability issues and greatly reduce points of failure. In addition, the investment in existing business logic is preserved since much of the existing code can be used by the Web implementation. However, the requirement to modify mainframe code to redirect the application I/O and to train mainframe programmers in Web techniques may mean this approach is better suited to new development.

The most interesting solution combines a mainframe Web server and a screen scraper. That is to say that it uses a two-tiered architecture to eliminate bottlenecks, yet requires no program modifications — the existing code remains untouched. This consists of a system that dynamically translates the 3270 data stream into HTML on the mainframe, yet allows the ultimate presentation of the Web pages to be under the control of a Web designer — a person with no mainframe experience at all, who is free to change the appearance of the presented screens in any way he or she wishes — meanwhile without affecting the application code.

The dynamically translated screens may be presented directly to the user without modification or they can be used as the basis of a template. To automate the process, a "snapshot" is taken of the translated screen and the resulting template is straight HTML that contains all the data fields of the original 3270 screen. Residing on the mainframe between the legacy application and the user, the template manager allows the Web designer full freedom to change the presentation — add graphics or multimedia, add, delete or move data fields — while the mainframe application still thinks it’s talking to a 3270.

This is an ideal solution for all involved — it just "drops in" and the legacy systems remain unchanged. Yet the application can be made attractive to the end users in a manner limited only by the imagination of the Web designer.

What about new development? Industry pundits are suggesting that the future of application development may lie with a trend called "Weblications" — the idea of building serious client/server applications that run inside a browser. Weblications represent the latest of the three enterprise computing models:

1. Dumb clients. (Mainframe-centric) All processing is done on mainframes.

2. Fat Clients. (Client-server) Most processing is done on the client and host is used as a data repository.

3. Universal clients. (Weblications) Browsers front-end the mainframe and most processing is done on the host.

Since Weblications don’t require a high-powered PC on every desk, they will be very cost-effective for companies to deploy. They also provide portability, flexibility and scalability because the Web is based on open standards. In addition, they are easy to install and update.

It is for new development that the mainframe Web server really begins to make sense — particularly if it is hosted in CICS. In this model, mainframe application developers code their business rules the same way they always have using proven techniques, but instead of coding for a 3270, their application I/O talks to the Web browser through a CICS-based Web server. In this implementation CICS data and applications can be accessed directly without the complexity of a gateway or middle tier.

A CICS Web server is the key to eliminating bottlenecks since it eliminates the gateway and allows the CGI script to work directly with CICS transactions. To further improve throughput, the CICS Web server can have an easy-to-use API to convert heavily used CGIs to an extension of the server, similar to the Netscape or IE APIs.

The enterprise computing model is shifting once again - this time to the universal client. Applications and their data are again becoming part of a centralized IT department. As part of this centralized strategy, mainframe sales are surging. End users will interact with their applications using Web browsers that are independent of platform or application.

Legacy systems may be accessed from an emulator within a browser or have a Web front-end spliced on. Emulator software can be used to deliver graphics and multimedia. End users with 3270s and a CICS signon can browse the Web.

CICS continues to be the most important platform for mission-critical data and applications. A CICS Web server is the one tool that can assuredly exploit the coming trends in application development. It is the one tool that can effectively leverage the years of investment in mainframe applications .