At most medium-to-large IT sites, applications and file services are spread out across the company on a networked quilt of Unix, Novell NetWare and Microsoft Windows NT servers. Monitoring and managing these applications and file services is a bear. Systems performance management does not occur in a vacuum. The load on any given system will change depending on the applications, processes, operating system, network traffic and user population. Shifts that occur day to day and week to week, as well as variations that follow business cycles, make server systems performance part of a continuum. This ongoing load variation, when applied to tens, hundreds or thousands of servers, requires specialized tools that can paint a perspective and diagnose these interrelated load changes.

Firing up a diagnostic utility such as Windows NT's PerfMon or the Unix System Activity Reporter (SAR) often is too little too late. It isn't that the memory, CPU, process, network, disk and performance metrics these products report are inaccurate. And it's not that they can't provide some raw data logging to give a limited historical perspective. The problem is that using such tools on their own segregates data collection by machine. On top of that, it's left entirely to the systems administrator to determine the data's significance.

To view the Report card.
What is needed are tools that identify and solve performance problems. The tools must monitor, diagnose, trend and even predict server performance. To this end we brought five systems performance management solutions into our Syracuse University Real-World Labs®--Boole and Babbage's Command/Post 4.1; Computer Associates International's Unicenter TNG 2.1; Hewlett-Packard Co.'s OpenView GlancePlus, MeasureWare, PerfView and Service Reporter; PLATINUM technology's ProVision ServerVision and Network Monitor 4.2; and Tivoli Systems' Enterprise Distributed Monitoring 3.6 and Tivoli Decision Support.

These tools struggle to deliver historical data that can be used to spot trends and predict future events and usage. CA, PLATINUM and Tivoli have improved their historical analysis. But HP, which captured our Editor's Choice award, offers powerful and intuitive trending, and its instant reporting raises the bar for this product category.

It's fair to say that these products are as complex as the problems they are meant to solve. We expected the implementation process to be tough, but we were pleasantly surprised to find all of the products are becoming better out-of-the-box product suites. It was apparent to us during testing that the vendors have learned some lessons about how to smooth the rough edges and deliver less painful solutions. Those lessons are translated differently with each product. For example, Boole and Babbage's Command/Post, unlike the other products, provides no diagnostics or trending but focuses on strong monitoring via alerts. CA's Unicenter TNG and Tivoli Enterprise Distributed Monitoring and Decision Support both run on Windows NT, which is generally thought to imply ease of use, especially compared with Unix. However, neither delivered real turnkey solutions. HP's and PLATINUM's experiences with performance management is obvious, giving their products a solution-oriented edge.

CA's Unicenter TNG was the most expensive solution, followed closely by PLATINUM's ProVision. This was no surprise: Both offer suites of enterprise management tools, including network and event management, as well as common services for storing data and messaging between distributed components. Tivoli, which is usually on the far side of ridiculous when it comes to pricing, managed to remain in the running with the best price and warranty. However, Tivoli Decision Support is promising but unremarkable in its current form. Also, while the price for Decision Support includes the Tivoli Management Framework, it does not include network or event management.