Tu ning NT: Is It Worth Your Time?

By Jay Milne  If money were no object, server tuning and optimization would be a lost art. Slow servers would be replaced with the newest and fastest servers available. But for many organizations, cost is a constraint and throwing hardware at the problem isn't an option. And upgrading hardware, such as processors and memory, can be more trouble than it's worth because of service interruptions. If you ask the people at Microsoft Corp., they will stick to their claim that Windows NT is a self-tuning operating system. Based on our testing, we'd have to agree. Other than adding memory, you won't find a magic bullet that will boost NT's performance more than a notch or two. If your server is running at low-to-medium ca pacity, the default NT install does an adequate job of tuning performance. In Network Computing's San Mateo, Calif., lab, we ran NT Server through its paces, focusing on adjustable operating system parameters and configurations that affect CPU utilization, memory utilization, disk I/O and network I/O. We investigated tuning techniques that make modifications via the NT user interface and registry. Following are the results of our testing as well as the benefits that were gained by the various tuning adjustments. Keep in mind that a complete analysis of your server should take into account the type of applications (such as file and print, database, Web server, groupware and so on) that you intend to run. The Test Bed To gather our quantitative data, we used a mix of eight 90-MHz Pentium and 200-MHz client systems running Windows NT 4.0 with Service Pack 3. Each client had between 64 and 128 MB of RAM. Our test server was a department-class Dell Com puter Corp. PowerEdge 4200, which ran NT with Service Pack 3 and Microsoft SQL Server 6.5--a typical setup for departmental environments with up to 250 users. Our server was configured with dual Pentium Pro processors (512-KB L2 cache), 256 MB of RAM and a 24-GB RAID array (six spindles). In addition, two external Fast SCSI II drives were attached via an Adaptec SCSI controller. All machines were connected to a single, isolated 100-Mbps Ethernet segment. We used Bluecurve's Dynameasure 1.5, a Windows-based load-analysis tool with eight clients, and a total of 50 motors (instances of clients) to evaluate our server. Dynameasure models both file service and database applications. We performed more than 40 tests and covered the majority of NT's core services. To monitor the server while testing, we used NT's built-in Performance Monitor as well as Network Associates' NetXray network-analysis software to measure network utilization.

Other Workshops Thwarting Malicious Java Attacks By Gregory Yerxa Related Links Is NT Strong Enough to Carry the Load? To Dupe NT Disks Or Not To Dupe? Multihoming And NT: A WINning Combination Remote Management of NT Servers Can NT Balance the Network Management Load? NT Remote-Access Se rvers: Ready For The Big League?