Microsoft's recent announcement that it will enter the embedded systems market with Windows NT surprised the dedicated OS industry. Microsoft had hinted that it might port NT to the embedded space, but industry insiders were doubtful of such a product ever materializing. After all, Microsoft already has an embedded OS—Windows CE.
However, with the announcement of NT Embedded 4.0, it's time to start asking some hard questions. For example, how will Microsoft differentiate NT Embedded from Windows CE? According to company officials, the differentiation is in the feature set and system footprint. In the case of features, Microsoft officials say NT Embedded will be the choice in situations requiring full Win32 support or BackOffice integration. If the situation calls for cramming as much functionality as possible onto a wallet PC device, the logical choice will be Windows CE.
Platform confusion isn't the only concern. Microsoft also must prove NT's suitability for embedded operation. NT is in a state of architectural flux. Major changes are on tap for the Windows 2000 (Win2K—formerly NT 5.0) kernel, and the changes add to the uncertainty surrounding NT's reliability and robustness. And nowhere are reliabilty and robustness more important than the embedded OS space: A failure in a critical system such as a router or LAN/WAN gateway could prove catastrophic.
Another matter is NT's console-centric management architecture. Although Microsoft is promising headless (i.e., no keyboard or mouse) functionality with NT Embedded, many Win32 components still present a dialog box on the server's console. This fact is especially true with Visual Basic (VB) applications. Errant dialogs can stop remotely managed NT servers in their tracks. Microsoft's ability to deal with these concerns will affect NT Embedded's success.
ANALYSIS
Despite the challenges, NT Embedded should be a welcome addition to the Microsoft OS lineup. Enterprise IS professionals will welcome the ability to leverage investments in Win32 programming knowledge and experience across a new class of systems. In addition, IS professionals will appreciate the simplified manageability that stems from a common computing platform. Still, Microsoft has to prove that NT is ready for key embedded systems roles. One market in particular that Microsoft has identified as a potential NT Embedded target is the one for medical equipment. The infamous phrase "blue screen of death" takes on a whole new meaning when mentioned within the context of a lifesaving medical device.
—Randall C. Kennedy
PC SERVER HARDWARE GROWS UP
Two crucial developments in PC server hardware will break barriers in 1999. The first development is the availability of 8-way symmetric multiprocessing (SMP) systems based on Intel's Profusion interconnect architecture. The second development is the move away from standard PCI-based I/O and toward more robust solutions, such as Intel's proposed Next Generation I/O (NGIO) for Servers.
Profusion will correct the problems of earlier 8-way solutions. Before Profusion, the best that you could expect from Windows NT-equipped 8-way database servers was average transactional throughput of 16,000 transactions per minute (tpm) as measured by the Transaction Processing Council (TPC). Limitations in the underlying interconnect architecture, which was mostly Axil Northbridge-based, and an emphasis on Pentium Pro-based designs were partly responsible for this performance.
The current generation of 4-way Pentium II Xeon processors had better results as measured by the TPC. Recently, HP's NetServer LXr 8000 broke through the 20,000tpm barrier for the first time under NT and SQL Server 7.0. If a 4-way Xeon system can perform at this level under NT 4.0, chances are good that we'll see throughput in the upper 20,000 range with these chips on the highly optimized (for Pentium II Xeon) Profusion architecture.
Of course, bigger boxes need more robust I/O. The NGIO model incorporates a new, fabric-based channel design that in theory will allow throughput of up to 10Gbps. NGIO also provides improved fault tolerance because devices are isolated and can't hang the bus the way PCI or ISA devices often do. Combined with Profusion, NGIO will help Intel-based servers challenge at least the midrange of the RISC server performance at price and performance levels that UNIX systems can't touch.
ANALYSIS
When NT-based servers fail to perform, most experts blame the OS. However, a lack of x86-based hardware scalability is also a factor. Now, with 4-way Xeon systems challenging the performance levels of Sun Microsystems' Enterprise and Ultra Enterprise boxes, you will be able to consider NT-based systems for high-throughput transactional applications. More important, these NT systems will approach UNIX-level performance for a fraction of the cost (as measured in dollars per tpm) of a traditional RISC system.
The future of NGIO, however, is harder to predict. Although Intel is committed to the specification, at least one major server hardware vendor, Compaq, has yet to endorse it. Compaq officials are leaning toward supporting the proprietary PCI-x architecture Compaq has been developing with HP and IBM. PCI-x will arrive earlier than NGIO—in the second half of 1999 vs. mid-2000 for the Intel specification--so PCI-x has an opportunity to establish itself as a bridge solution between current PCI designs and NGIO. Still, NGIO isn't going away. Switched fabric holds a performance and reliability advantage over the traditional PC bus.
—Randall C. Kennedy