Pinpointing Unix performance problems is a pain in the net - Unix performance monitoring software

Software Magazine, April, 1995 by Paul Korzeniowski

In the past, a typical IS response to Unix performance problems was to reboot the system or, more seriously, upgrade hardware. As Unix installations grow, administrators look for tools that provide both historical trending and realtime data.

Unix beckons IS people with its promises of openness and flexibility. As shops begin to trust important applications to the Unix platform, they are looking for tools to manage their new systems.

One of the things they want to know is whether their new systems are performing to capacity. Alas, managing performance for distributed Unix systems is much more complex than it was in the land of the mainframe, where tools measured such details as CPU utilization so companies could be sure they were squeezing every possible dollar out of their big investments. Consequently, the tools available for monitoring Unix performance are what many would call "immature."

A few tools have been available for a couple of years, but several are still in the beta test stage. Suppliers who want to capture market share must answer a number of questions: How similar is Unix performance monitoring to mainframe monitoring? Do customers want links to network management platforms such as Hewlett-Packard's OpenView, IBM's NetView for AIX, and SunConnect's SunNet Manager? How important is DBMS monitoring?

At this point, no one knows the answers to these questions. Exactly what issues need to be addressed for Unix systems will likely become apparent as vendors enhance existing tools and deliver new ones.

The cloudiness surrounding Unix performance issues has resulted in less than spectacular interest in the current crop of offerings. Even performance monitoring tools suppliers admit that Unix customer interest has so far been tepid. "Performance monitoring is like a VCR," explained Steven Johnson, a marketing manager at Landmark Systems Corp., Vienna, Va. "If you do not have one, you may not be sure why you would need it. Many Unix users are just beginning to understand why they need performance monitors."

The bulk of companies who have shown interest in Unix performance monitoring are those familiar with mainframe monitoring tools. One such company is Chemical Residential Mortgage Corp., Worthington, Ohio, which moved an automated work management application from its IBM mainframe to Sun Sparcstations in the summer of 1993.

While the company made the switch, Randy Keefer, the firm's manager of operations and technical support, searched for performance monitoring tools. According to Keefer, few tools were available.

Despite this, the firm downsized the application and used Unix utilities to ensure adequate performance. The results have been mixed. Keefer said the realtime monitoring tools have been adequate, but he has found it difficult to obtain historical information.

Keefer's desire for management tool products that detail both realtime and historical data is a common one. Systems administrators want tools that measure actual response time and illustrate problem spots. They also desire tools that store that information so they can tinker with it to get a better understanding of how components are functioning. Some products feature both capabilities; others support only one.

Although they had no tools to handle historical data, Chemical Residential Mortgage rolled out the work management application to 450 of its 900 employees. They have stayed on top of performance problems by estimating how much memory and network bandwidth each additional user requires, said Keefer.

The firm plans to serve as a beta test site for the Resource Optimization Control (ROCS) line from Legent Corp., Herndon, Va. Announced in September 1994, Legent's Paramount/XP ROCSview monitors performance on mainframe, IBM AS/400 and Unix workstation platforms.

Unified Interface Sought

To boost its own ROCS sales efforts, Legent inked a reseller agreement in late 1993 with Hewlett-Packard (HP) Co., Cupertino, Calif. Legent linked its performance technology with HP's Performance Collection Software (PCS) interface.

For its part, HP is working with 12 other systems management suppliers, including 4th Dimension Software Inc., Irvine, Calif.; SAS Institute Inc., Cary, N.C.; and Systar Inc., Reston, Va., to improve customers' performance data collection.

The program seems to be paying dividends for HP and its partners in the form of some large customers. Charles Schwab & Co., San Francisco, has 6,000 users who work with mainframes, Unix workstations and PCs. In the spring of 1994, the firm started to downsize mainframe applications to LANs connecting HP, IBM and Sun workstations running Unix and Microsoft's Windows NT.

In its search for a performance monitoring tool, the brokerage house found the pickings less than plentiful. "When we were looking for a monitoring tool we talked to a lot of vendors, but only a few were actually delivering products," said Ron Welf, manager of performance and capacity planning. They examined products from HP, Landmark, Legent, Digital Equipment Corp. (DEC), Maynard, Mass., and OpenVision, Pleasanton, Calif. Welf said they selected the Legent package because it could be integrated with HP's OpenView.