Performance monitoring, consider what's (not) happening

The classic monitoring case focuses on particular data items exceeding thresholds. Simple things like:
  • CPU is > 75%
  • Average transaction response time > 5 seconds
These measures do, of course have value, but it can be helpful to look at what is not happening as well. Lately I have had several customers talk to me about how they watch for lower than normal workloads or conversions in their funnels as leading indicators that something is wrong.
While “less than normal” behavior evaluations work at the infrastructure layer, they get really interesting when applied to transactional behavior. And, doing so often challenges IT to think differently in the process. For example, seeing reduced transaction conversions might be a result of poor performance that needs to be checked (in which case you will probably find some complementary system behaviors which are causing it), but it might be also be a clue that some recent application change made the site less usable.
Making these kinds of discoveries elevates the value of IT to the business, and is one simple example of how applying application performance monitoring strategies can help IT and business teams align.