I was listening to the radio this morning, and they announced some "fun" statistics in the way that they often do to entertain. For example;
- The average person spends 3 years of their life driving their car
- The average woman spends 2 years getting ready, of which 8 months is spent putting on make-up
- The average man spends 1.5 years cooking.
This got me thinking about the absurdity of using these averages for anything apart from entertainment. For example, it would be wrong for a man to cook like mad for 1.5 years when he is 20 and then refuse to cook another meal for the rest of his life. Equally, if you don't pass your driving test, how will you ever rack up the requisite 3 years driving your car? These "values" are not focussed events that can occur in a single batch, but then again neither are they regular events that can be spread evenly throughout your life (if you live to be 70 years old, you can't expect to spend 1 hour each day driving... especially since you couldn't reach the pedals until you were 11!
Average values have their place... treat them without intelligence and you are playing with meaningless (and useless) information.
What has this got to do with Capacity Management?
A common "heated discussion" that is had with some IT technicians relates to the use of an average CPU utilisation. It is claimed that averaging the CPU utilisation is a meaningless thing to do.
That all depends on when you do it, and for how long. Averaging the CPU utilisation over a whole day is unlikely to be useful for a Capacity Manager given the amount of variation in activity that one would expect to occur during that day. However, you have to use some sort of averaging. Think about it. A CPU is either doing an instruction, or it is not. If you could monitor a CPU at the lowest possible level, then you would see it either busy (100%) or idle (0%). You couldn't do any management at this level of granularity, so it makes sense to use an average. But at what interval?
Some workloads are very quick, and therefore averaging the CPU utilisation at 5 or 10 second intervals might be required. It might still give a rough-and-ready indicator of the amount of work done, but measuring at a lower level is likely to be impractical.
Some other workloads may take many hours to complete. Measuring CPU utilisation at 10 minute or even 1 hour intervals would be perfectly adequate in this scenario.
Fundamentally choosing the timeframe over which to measure your averages is just as much a skill of the Capacity Manager as knowing what to do with the values once you have got them.
There. That is another 20 minutes of my 6 years of Internet use gone!