Measuring the performance of an HP 3000 used to be a leverage point for increasing investments. By now the numbers help justify continuing to use the server in a datacenter with newer boxes. "We think of our HP 3000s as stable, and even reducing in usage over time," says one systems manager, "though actually as the company grows, the data requirements and load on the 3000s increases."

One way to measure a 3000's footprint is the amount of memory it requires. Memory upgrades cost nothing like what they did even 15 years ago. But any spending at all makes that 15-year-old server suspect. HP's Steve Macsisak recommended sessions x 4, plus jobs x 16, plus 64 MB as the criteria for memory usage.

An HP 3000 uses as much of its memory as possible to make processing efficient. The design of the PA-RISC architecture makes memory the most important element of performance, after IO speed. It's not that unusual to see a 3000 using 100 percent of its memory, according to field reports. There's also CPU usage to measure. 

CPU percentages can come via the REPORT command. Count up the CPU seconds used in the week, and divide by the total number of seconds available (604,800). But for all of this, it doesn't feel like a graphic report the rest of the datacenter gets from its Unix and Linux systems using SAR. There may be a program inside a 3000 that can help, even if the company never purchased performance tools from Lund. HP's Glance gives away its reporting power in its name, one manager has joked.

There's freeware available to create handsome graphs like the one at left, suitable for showing in a meeting about datacenter resources. Ploticus/iX was written by Andreas Schmidt. It uses data from SCOPE.SYS. Ploticus even works with SAR's data.