The Big Trend -- Unix, Linux and Windows Will Fight it Out for Years
- By Scott Bekker
As they always do in IT, the rules are changing when it comes to which operating system belongs in which role in the enterprise. The old formula was Unix on the high-end application servers and database servers; Windows in the networking servers, file servers, print servers and e-mail servers; and Linux on the front-end and Web servers. The current trend away from Unix/RISC toward more economical x86 architectures opens the door for all-out war between Linux and Windows for a bigger chunk of the server market. But you can bet that Unix will not go without a fight and may never disappear.
In fact, the picture is always much more complicated with the different platforms occupying positions in all three tiers of an enterprise. Any individual shop may be mostly Unix or mostly Windows in the server room. Inside a shop, the ERP application may have a Unix back-end, while the CRM application might have a Windows back-end. An individual department may be running Linux file and print servers even though the corporate standard is Windows. Many of the moves away from Unix are occurring in those odd areas. But the overall trends are occurring in the context of the major divisions of labor.
What analysts and vendors see happening now is different from what the biggest boosters on each side would prefer. Many Linux purists would like to see Linux eat into Microsoft's market share. Many Microsoft employees would prefer to see Windows work its way into the highest reaches of Unix deployments. Neither side is getting its way. Meanwhile, some observers think that the way operating systems are selected is changing in a fundamental way that will make the operating system itself less and less important.
The Big Trends
Unix is being chased upmarket. Whether you ask IDC, Dell, IBM or Microsoft, the key driver is the economics of x86-based architectures. The improving performance of both Linux and Windows servers on standard Intel-based hardware makes it difficult to justify the premium pricing of RISC-based servers. Economic conditions have helped crystallize this thinking for IT departments -- whereas Unix enthusiasts within organizations might have held sway with technical arguments when budgets were big, acquisition cost grew in importance in recent years. The impact on server vendors has been profound. Sun Microsystems is struggling to define its value. SCO has taken to suing over Linux and intellectual property. HP and IBM have been protected by their strategies of offering whatever platform a customer wants. HP also laid strategic plans to co-opt the issue by migrating its HP-UX flavor of Unix to the Itanium processor, undercutting much of the economic argument to switch applications from Unix.
Linux is doing most of the chasing. Linux is chasing Unix out of the low-end from the Web servers that used to be run by Sun's "dot" in "dot-com" to the lower tier infrastructure and low-end application servers that also used to be Unix. Analysts like Al Gillen at IDC credit the portability of skills and applications from Unix to Linux as key to Linux' success in those areas.
But Unix is holding Microsoft off at the top of the market. The performance gap between the largest Windows servers and the largest Unix servers is almost gone by any scalability measure. Windows servers now support up to 64 processors, all 64-bit, and 512 GB of system memory. On large-scale benchmarks, like the TPC-C, Windows and Unix results are now practically indistinguishable -- Unix currently retains a slight performance edge, while Windows retains the cost advantage. But Windows still lacks the mindshare among the IT buyers considering large systems. A mindset still exists that Windows remains in the basement of scalability despite evidence to the contrary, and many users even labor under the mistaken impression that Linux is more scalable in large SMP systems. It is a mindset that Microsoft is working to change, but the Catch-22 is that Microsoft must get more reference customers onto large-scale Windows databases and application servers. To be sure, deployment cycles for massive, mission-critical systems are much longer -- and Microsoft's newfound scale-up capability has only been shipping since April with the launch of Windows Server 2003. The time is coming for this shift, but it's not here yet.
Scale-out is growing in importance, and Linux is benefiting the most. Despite Microsoft being the evangelist of scale-out from the year 2000 until recently, it is Linux that appears to be benefiting most from the emergence of that trend. While Microsoft has a high-profile implementation on Dell hardware at the Cornell Theory Center, it is Linux that dominates this field. Dell, which has promoted high-performance computing clusters heavily, estimates that 90 percent of its HPCC implementations are on Linux. So far, HPCC is a tiny percentage of the overall market, but the clusters are becoming an important segment of the high-profile Top 500 supercomputers list.
Windows remains strong in networking, e-mail, file and print serving. Windows Server 2003 enhanced the strength of Windows in all those core areas, and much of the operating system's growth will come from new systems brought in by enterprises to replace their aging Windows NT infrastructures.
Windows is the default app server at the sub-mission critical level. Most major new systems going into the enterprise are going in on Windows servers, and Microsoft enhanced its application server and development tools significantly enough in the 2003 generation, to make that continue to be the case. "We think Windows is going to continue to be the application server of choice when it comes to anything but the highest-end implementations," says IDC's Gillen.
Web servers continue to proliferate, and Linux retains an edge. Two recent tidbits from Netcraft, the U.K.-based outfit that regularly surveys the Web for trends in Web server usage. One is that Apache recently hit a record share of the Web server universe, and that remains well over 60 percent. The other is that Windows Server 2003 was headed for 100,000 Web server deployment in no time flat. Conclusions from those Netcraft tidbits: The open source Apache Web server, normally run on Linux, is the Internet standard; while businesses continue to pay Microsoft for enterprise-related secure transactions and Web-based applications. One other development that is too recent to classify is the Web Edition of Windows Server 2003. Priced at $400, Microsoft hopes to compete better with Linux' much lower price point in the Web farm. It's too soon to tell if the price is low enough to make a difference.
Operating system choice may be less important than it used to be. With scalability limitations disappearing from the Windows side, Unix systems that run on Intel-based hardware and a growing realization that the real costs of systems aren't reflected in acquisition costs, more and more decisions may be made on the choice of a deeper platform. For the last several years, that platform choice has been developing into a two-way decision between the J2EE platform and Microsoft's .NET architecture.
Scott Bekker is editor in chief of Redmond Channel Partner magazine.