By Jean S. Bozman
Unix and Windows NT. I might as well have said oil and water. For years, there have been people advocating both types of operating systems, and they have clearly taken sides on which one is best.
Journalists have positioned the struggle as Unix versus NT for years now, and that reflects the marketing wars between various systems vendors. Next year, those marketing wars will heat up to new levels of intensity when a flood of business servers based on the Intel Pentium Pro processors hits the market.
Unix and NT will be battling it out in a big way on the same Intel hardware platforms; users will have their pick of operating systems when they buy the hardware. (The head-to-head Unix/NT competition on Intel really started several years back with SCO's OpenServer, Novell's UnixWare and Sun's Solaris for Intel.) And, by the way, there will still be plenty of Unix RISC servers out there, this time with 64-bit hardware, increasingly aimed at high-end performance.
So much for the marketing wars. Let's try to look at it from a user's point of view. Users in large corporations will find themselves surrounded by hundreds of distributed servers, some based on various flavors of Unix and some based on Windows NT. From a pragmatic point of view, these users will have to make themselves at home with that situation.
The NT server ramp-up will occur so quickly that next year IS managers will likely start asking some important management questions they might have asked sooner. Until recently, there has been a lot of focus on Windows clients accessing Unix database servers. That has been worked out, by and large. Now we'll have to get NT servers, which are built to host Windows applications in a Windows world, to "peer" with Unix servers, exchanging data with them as easily as they would with another NT server. That will require some homework--and a lot of new connectivity products.
Three downstream factors will drive the choice of operating system on the Intel platforms: the types of application, database usage and Internet usage planned in a department or division. In some cases, NT will be favored for its affinity with packaged Microsoft office applications. And it will win where IS managers perceive that NT on Intel is a more cost-effective platform for business applications that must be deployed across a large company.
But Unix will continue to outpace NT in terms of the number of processors supported, the scalability of existing applications and the size of databases supported. That means that scalability under one copy of the operating system--for example, a single symmetric multiprocessing (SMP) server--will be able to reach higher under a Unix operating system for some time. However, NT aficionados would note that an era of clustered NT systems is about to begin, allowing groups of Intel/NT systems to cooperate on large workloads.
Let's accept the fact that many organizations will own both Unix and NT systems. When you have an IS infrastructure made up of dozens or hundreds of distributed servers on a universal network, business units can buy application packages written for either or both OS, or rearrange their computing environment at will.
Once those dozens--or hundreds--of servers are in place, there is a lot of hard work to do to allow Windows NT and Unix servers to coexist easily with each other. Still missing, for the most part, are the network management, systems management, database and cross-platform applications development tools to make these links seamless.
What can be done to bridge the gaps that currently exist? There are software tools to port applications from Unix to NT or vice versa. Both operating systems support development in C, C++ and Java, among other languages. To the extent that users deploy corporate applications to both platforms, the difficulty of maintaining a mixed-vendor environment will be minimized.
Database systems have already been ported to run on both types of systems, easing NT-server-to-Unix-server database queries. Existing technology from Oracle, Sybase, Informix and others can be leveraged to make querying NT and Unix servers a more user-friendly experience.
But what if a user wants to access data from both environments? Internet technology may come to the rescue in several ways. First, the universal HTML language of Internet browsers, including those from Netscape, Microsoft and Oracle, will allow universal client access to NT and Unix servers. A variety of platforms can still run a variety of operating systems, but that need not hamper client/server access through a standard set of Internet browsers installed company-wide.
Second, Internet applications will allow users to attach those universal clients to a variety of Internet or intranet servers, as well as legacy databases and legacy machines. Prime among these will be Internet front-end applications, written in C++ or Java, that connect users' Internet browsers with existing applications and databases. This is a step forward from writing object-oriented "wrappers" around legacy application code.
As to connectivity, the Internet/ intranet phenomenon is forcing Microsoft and the various Unix vendors to gravitate to a common set of Internet and networking standards based on TCP/IP. That will provide a level playing field for all the vendors, transforming point-to-point client/server applications into distributed, transaction-based information systems. But it will do so without scaring developers with the notion of object-oriented technology that is so distributed that it is out of control.
On another level, the NT and Unix servers must be managed from the same set of corporate IS consoles, or business operations will suffer. Systems vendors will step up with middleware components and systems management tools that link many NT servers with many Unix servers. In the process, Unix system management tools will be ported to run on NT servers.
Microsoft looks to systems vendors, including Hewlett-Packard and Digital Equipment Corp., to lay much of the groundwork here. HP and Digital have already started down that path, leveraging existing systems management software for the task of controlling mixed Unix/NT networks.
HP and Digital have the motivation to do so: Both sell NT/Intel and Unix RISC server product lines, as does IBM. But others will sell Unix and NT on the same Intel systems, among them Data General, NCR, Siemens-Nixdorf and Unisys. Still not spoken for are Sun Microsystems and Silicon Graphics, which sell Unix systems only, although Sun sells its Solaris Unix for Intel platforms.
Microsoft is encouraging independent software vendors (ISVs) to create and sell NT/Unix connectivity software, such as AT&T's Advanced Server for Unix, which allows NT servers and Unix servers to act as peers on a network. Smaller ISVs have NT-to-Unix connectivity software, too.
This scenario presumes that it's in the enlightened self-interest of users, vendors, VARs and ISVs alike to have a two-system universe. It presupposes that application developers and system administrators will acquire skills that work in both environments. And it supposes that each environment will play to a set of strengths that the other does not.
If this mixed-vendor scenario plays out, the ultimate winner is the user in a large enterprise--as well as Internet users at home and in small offices who are accessing distributed servers. Without cross-platform development tools, applications and databases, the world of distributed servers would be chaotic. Users already got a taste of that in the world of client/server computing that emerged in the 1990s.
Luckily for users, a wave of Internet technology is sweeping over information systems in 1996, blanketing all clients and servers with a new layer of connectivity software and masking the differences between desktop clients. That will go a long way toward making truly distributed computing more tolerable in the late 1990s.
Jean S. Bozman is research manager for Unix and server operating environments at International Data Corp. in Mountain View, CA.
Did something in this column press one of your hot buttons? Then let us hear what you think by sending a response to pubs@uniforum.org. We'll consider it for publication in "Letters to the Editor."