Reliability, availability and Windows 2000.
Microsoft Corp. would have us believe these three words were destined to go together like, “Baseball, apple pie and Chevrolet.” In Chevy’s case, however, I don’t think General Motors’ appeal made any difference to the company’s bottom line – customers made their purchase decisions based on performance results. If Microsoft’s destiny is determined in the same way, it could be facing an uphill battle.
Windows NT users, with blue screens burned into their retinas, applauded Microsoft’s emphasis on reliability with Windows 2000. Finally, all the bubble gum and duct tape that bandaged Windows NT together with various service packs and patches would be removed, and users would have a clean, shiny, new, blue-screen-free operating system.
Unfortunately, those of us who have peeled the wrappings off the new package have discovered residue from the NT bubble gum and duct tape, along with a few new wrinkles. Just scratching the surface of the mountain of Win 2000 reliability features recently resulted in a disappointing case of deja vu.
In Tolly Research experiments assessing the capability of server-port aggregation software to increase Fast Ethernet throughput to Win 2000 servers, engineers discovered that modest traffic loads (three or more simultaneous file uploads) brought the server to its knees. When configured with Intel PRO/100 Server Adapters and running Intel’s Adaptive Load Balancing software, we observed blue screens galore.
Webster’s dictionary defines reliability as the, “extent to which an experiment, test, or measuring procedure yields the same results on repeated trials.” In that regard, the server-port aggregation was extremely reliable – it crashed in every trial. That wasn’t after 10 minutes of nonstop uploads – it crashed immediately.
In its knowledge base, Microsoft does acknowledge that, “Under certain circumstances when you are using TCP/IP for network communication, your computer may stop responding [hang] . . . Microsoft has confirmed this to be a problem. . . .”
This acknowledgement undoubtedly will give you a great deal of solace as you repeatedly reboot your server.
In fairness to Microsoft, we did not observe the same catastrophic result when using 3Com network interface cards and software on a Win 2000 Server. However, we couldn’t detect any bandwidth aggregation, either. I guess you have to pick your poison.
Microsoft points to a number of recent studies regarding Win 2000 reliability, but they almost exclusively focus on Win 2000 Professional, not Windows Advanced Server, which is supposed to support strategic mission-critical applications. While desktop reliability is a noble goal, server reliability is of far greater significance.
Because of the requisite interplay of the network operating system with third-party hardware and software, ensuring server reliability can be a daunting task. It is incumbent upon Microsoft, as the end-to-end solution provider and guarantor of “reliability,” to bring about dependable interoperability.
Microsoft, no doubt, will chip away at pesky Win 2000 idiosyncrasies (like server crashes) until they legitimately can proclaim high reliability. The question facing those considering the transition to Win 2000 Server is, “When?” Early indications are the answer is, “Not yet.”
Flood is chief technology officer of Tolly Research. He can be reached at