Tech Tip

By Paul Murphy, author of The Unix Guide to Defenestration

There's a management maxim: "you get what you measure" that pertains to human performance in large organizations. What it means is that people will generally maximize performance with respect to whatever parameters you measure them on while, of course, minimizing others.

As a sysadmin you can't be effective without knowing a lot about how your systems are peforming and you've probably graduated from Adrian Cockcroft's original proctool for Solaris 2.5.1 to SE/Symon and may now be looking at DTrace for help.

DTrace, in particular, is both brand new and the first real step forward in resource use debugging in a very long time, but all of these tools have two things in common:

  1. First, they are oriented to helping you figure out where to focus your attention with respect to your role as performance management czar on your gear; and,

  2. Secondly, they don't help you much in communicating your findings to whoever controls your hardware budget.

When it comes to budgets, measuring to get what you want means communicating the results of your measurement activities to people who don't understand the hardware, the load, or the parameters being measured.

Very often the right answer to this comes in the form of simple color charts. A written statement like: "between 8:15 and 12:00 we typically run at 63% average CPU utilization on the Oracle Server with an average of six incidents in which CPU hits 100% for periods of three minutes or more - thus delaying users" may be accurate, but it's not going to loosen the purse strings if they're held by non technical management.

What will usually work is a combination of user complaint about the delay coupled with a nice color chart showing the CPU maxed out for minutes at a time and a recommendation on how a few bucks can solve the problem. It's the same information, but it's the packaging that makes it credible.

ALmost anything will do to make these graphs -as long as they look pretty and don't lead Finance to assume, as they will if you use Excel, that you're doctoring the numbers. COmmercial products like sarcheck and glance have the right graphs, but even something as simple as screen dumps made using perfmeter under CDE and Solaris 9 will work quite well.

What's going on here is that people in departments like Finance almost always treat hardware budget requests coming from IT as self serving and thus subject to automatic rejection. Wrap that request in a user complaint, however, and the presumption of self interest affecting anything you write disappears. Now, instead of looking for a reason to reject the request, most Finance people will look for a reason to accept it.

Your color chart may be over simplified and silly, but that doesn't matter. What does matter is that you get what you measure -provided someone else presents it.