View from the Hill - #4

For large systems, Solaris on SPARC will continue to be a better choice than NT for many years.

Software publishers have many common attributes. We parodied some of these in our Spellerbyte software page. But there are some serious underlying assumptions which users would recognise from their own experience.

• The reliability of an operating system tends to increase with age. Within any particular OS, the later you start to use it, the more likely it is that someone else has encountered and fixed most of the bugs. Provided that the OS you are looking at includes the functionality you need, then the older the OS, the more reliable it is going to be.

• Application support increases with the age of the OS and the size of the user base. Third party software publishers have limited resources, and if they support a new OS too early they risk supporting a dud. If they support new features too late, they lose market share. But unless they are supported by venture capital, or cash cow products from other platforms, it's generally safer for them to wait for new features within an OS to get established.

• Software is written by human beings, who need to get up to speed with new OS features along with everyone else.

I remember that when Sun Microsystems announced its first multiprocessor server, the SPARCserver 600MP in 1991, my colleagues and I were both pleased and amused at the same time.

We were pleased, because we had been building multiprocessor systems for real-time applications for some time. The task of integrating a whole bunch of disparate processors and operating systems (one optimised for database, another for real-time response, and yet another for number crunching) was good fun, but it consumed a lot of engineering effort and architectural design. In the systems we built, the Sun was the cheapest component, and we were already starting to interconnect SPARC systems across the VMEbus. This looked like a good way of saving money for our customers.

We were amused, because we knew from our own experience that once you created an architecture like this, it could take several years before you got the full benefits. We speculated that the only benefits you might get straightaway would be if one processor was managing the display while another was managing the disk drives. This looked like an expensive option compared to intelligent I/O and graphics accelerators, but nevertheless one which would offer great flexibility once Sun started to fix the lack of real-time support in SunOS.

In 1992, Sun changed the face of computing forever by shipping the SPARCstation 10. This was the first time that you could get a low cost high volume box which would run real applications, and included multiprocessor capability. From that point the clock started ticking for a new software market. Previous multiprocessor systems were unattractive to most applications software publishers, because they were either:-

• low cost and without an operating system (such as the transputer or various array processors intended for DSP applications, or

• ran an operating system (such as Cray or IBM scientific computer) but cost millions of dollars and had a total population of tens or hundreds of any particular machine.

It was 1997, before similar types of systems started to appear in the Intel PC market, and the methods of interconnecting the various processors within a system were implemented differently by different manufacturers. It was probably not until 1998 that any kind of multi vendor standardisation started to appear in mulitiprocessor PC systems, and that's when the clock really started ticking for third party software vendors who wanted to use these features in a PC running NT.

My own view of the Solaris versus NT debate for large scalable systems is based on the gut feel notion that the real difference doesn't lie so much within the technical capabilities of Sun Microsystems or Microsoft, who are equally capable of recruiting and deploying good software engineers. The real differences that users experience today, are more deeply rooted in the longevity and applications experience of the third party applications software developers. In this race, Solaris applications had a clear 5 year lead. That doesn't mean to say that NT can't catch up in some areas, or even overtake Solaris.

But if you believe that it takes a couple of years before new features in an OS are used efficiently by third party developers, and then it takes another couple of years before the applications bugs are ironed out, you can appreciate why in the year 2000, and maybe awhile yet, the question which is better, may be answerered by the one which has matured longer.

Readers who seek more analytical discussions on this subject are referred to the articles linked at the top of this page.

SPARC(R) is a registered trademark of SPARC International, Inc. SPARC PRODUCT DIRECTORY(SM) is a service mark of SPARC International, Inc used under license by ACSL. Products using the SPARC trademarks are based on an architecture developed by Sun Microsystems, Inc.