e i g h t - c u b e d . c o m 

Publishing performance stats

Since I got back to work, I have been heavily focused on investigating some performance issues that the users have been complaining about.

The structure of this particular application is a PC fat client talking to ACMS servers via ACMSDI. The application is load-balanced across two machines that both run ACMS and ACMSDI (in other words, the application is load-balanced twice, once when the PC connects to ACMSDI, and also when a transaction is requested of ACMS).

For some reason, there is a perception that one machine is faster than the other.

As we run Polycenter Performance Advisor, I headed straight for the graphs section of the reporting tool. Now, normally when we experience a performance problem, it is caused by someone doing something inadvisable (usually someone attempting to troubleshoot a production problem). They don't mean to cause an issue, but they do something like forget to set their transaction type to readonly, and we're screwed.

These types of issues are generally extremely easy to spot in Advisor's graphs, as they tend to cause spikes or dips at exactly the time the issue was reported. In this case however, nothing stood out.

When I reported this to the people involved, there was much disbelief. So I broke one of my rules and published the raw performance graphs.

I have this rule because over the years I have found that handing graphs to people that have little understanding of the computing performance model tends to generate more questions than it answers. For example, a graph showing that a CPU is 100% busy causes consternation until it is pointed out that 60% of the time is being spent running batch jobs at reduced priority. Unfortunately, the graphs don't show job priority, and hence the question.

What really frustrates me is that attempts to educate people about how to interpret the graphs generally result in responses such as "Why should I have to know that? That's your job." Yes, I know it's my job. So why not believe my verbal summary that the graphs show nothing and insist on me publishing them?

So far, this group of people have been pretty good about accepting interpretations. I can only hope that that will continue...