June 2001

"NT doesn't scale well enough to handle multiple applications.... so every application has to have its own machine. We're up to our necks in servers!" -- Stephen Litchfield, Network Computing, November 27, 2000, page 47.

I once interviewed for a networking position with a large manufacturing and sales company in southern Georgia. The lady in charge was concerned that her seven OS/2 servers were going to have to be replaced by Windows NT machines, and she was looking for a qualified, reliable techie who could perform the transition. She rightly insisted on hiring someone with a knowledge of OS/2 and experience with the platform.

Her concerns turned to fear and disgust when I gave her the bad news. Those seven OS/2 servers, which were used to provide applications and storage for some 400 workstations, were not going to be consolidated by migrating to NT. No, it was far worse: she would soon be responsible for some 20 to 50 NT servers. "More servers??!" She just about freaked out. She wanted to buy "one good computer" and just run all of the new server programs on that one computer. Leaving aside the dangers of using an error-prone NT system with "all of the eggs in one basket" (meaning that a single crash could obliterate her entire information infrastructure), the fact is that her goal was simply impossible to achieve. NT is not scaleable.

When we use the word "scaleability," what do we mean? The Microsoft definition means "you can add more servers any time you need to." The IBM definition is "you don't need to." It's really that simple.

With OS/2 Warp as her standard networking platform, the factory manager did not need to add new servers every time she hired a few dozen workers to scale up her factory volume and boost her sales force. She could have chosen to co-locate the applications on just two or three servers, consolidating her system and reducing hardware costs and networking complexity. This is because an OS/2 server can run hundreds of PC workstation tasks simultaneously, including multiple applications that don't have to fight with Microsoft's Neanderthal registry system. Both from a performance-curve view as well as an application-coexistence view, OS/2 is simply the superior choice for GUI-based, low-cost server operating systems.

Then why didn't the factory manager just stay with the proven, reliable performance of the OS/2 configuration? It was because her application provider refused to continue to maintain the native OS/2 application. She was being forced to downgrade to a Windows environment because of desktop software issues, not server performance or reliability issues. This is also known as "the tail wagging the dog."

Yes, a network manager is supposed to focus on goals such as zero downtime, improved speed, larger capacity, minimal conflicts, and consolidation of resources. Also included on the agenda are issues of security, expansion, and data integrity and preservation. A network manager should not have to waste time and resources because a key application is suddenly changed to a different platform. This is like having to buy a new house because the telephone company decided to change the type of telephone lines, or having to buy a new car because somebody cornered the market for spark plugs and discontinued your car's plugs.

For those who are unaware of the scaleability limitations of obsolete, legacy environments such as WindowsNT and Windows2000, it is best to simply let Microsoft speak for itself. A so-called "heavy user" scenario -- involving "full-page documents using Microsoft Word" and "even heavily browsing the company Intranet" -- on an NT-based Windows Terminal Server network requires the following configuration: dual-processor Pentium II with 512MB RAM, and only 15 to 25 users per processor recommended. This information is provided by Microsoft in a technical sales brochure; there are more details at

Meanwhile, a typical OS/2 system can run more applications simultaneously (due to far superior multithreading and the absence of Microsoft's legacy Registry obstacle-course). At the same time, the number of users per processor is typically over 50 for an entry-level OS/2 server and up to a thousand or more for the higher-end OS/2 servers. To get close to that kind of scaleability from Microsoft requires an ultra-expensive Windows Datacenter system -- not exactly the kind of investment that can be recouped before Microsoft declares the product obsolete and dictates another round of so-called "upgrades." And OS/2 has been the scaleability and price/performance leader for almost a decade already.

It is really absurd that software applications dictate the network environment. Solely due to the forced migration to a brain-dead Windows platform, the network manager now has to deal with ten times as many pieces of hardware, tons of new networking connections, a far more complex system configuration, multiple backups, hot-swap cutover systems, additional technicians on staff, more software conflicts, and numerous other emergencies. This is certainly not an upgrade -- it is a downgrade. No wonder NT sales figures look so good -- with Windows you have to buy up to ten times as many copies per project as a Unix, Novell, or OS/2 environment.

I remember the old picture from the 1980's -- the Novell era -- of a typical network manager. Full-bearded men with huge paunches and pale complexions, the typical image was that of a person who merely sat in front of a single computer screen and played God with people's access accounts. That was because they didn't have to get out of their seat to install new servers, to format desktop hard drives for reloading Windows, or to chase down arcane application conflicts on multiple servers. The new image is now a skinny nerd with a host of training certificates working 80-hour weeks and popping pills or slugging down Jolt Cola just to stay awake. This is not an upgraded life, but a downgraded one. No wonder demands for Windows serfs are so high -- with Windows you need to hire up to ten times as many bug-fixers per project as a Unix, Novell, or OS/2 environment.

The goal of creating applications that are OS-independent ought to be a no-brainer, but there seems to be few companies willing to dedicate themselves to large-scale applications for non-Windows platforms. Unix is the only real exception here, and this is because of Unix's historically high cost (and correspondingly greater revenues per sale). Until several software makers with large market-shares commit to cross-platform development or Java, the tail will continue to wag the dog. There are plenty of unscrupulous software consultants who gloat and salivate at the thought of selling into a market driven by lack of scaleability, requiring more and more investment by business owners. OS/2 has the fundamental qualities necessary to come to the rescue of desperate businesspeople. But just like the Y2K issue (see quote of the month above), sly consultants have a habit of recommending the most expensive solution, not one based on excellence.

Most recent revision: May 28, 2001
Copyright © 2001, Tom Nadeau
All Rights Reserved.