SQLServerCentral Article

A Grain of Salt


A True Statistic

I heard someone mention this in a seminar recently and it immediately woke me up. A good thing since I was starting to nod off. He was talking about team and change in IT, mostly systems, but also people, culture, etc. The statistic was that 50% of all changes fail.

Now I'm not going to debate that or even repeat the supporting arguments. That wasn't what woke me up. The thing that woke me up was a deep seated and barely contained urge to bray like a donkey at the absurdity of the statement. I limited myself to a smile that was probably misinterpreted as interest and belief.

You see, over the years I've worked with lots of numbers. Many many or them in school and work, some financial, some just data about a system of some sort that I had to interpret and work with. And having reported on data many times, I've grown skeptical as I've learned how numbers are manipulated to generate statistics, especially in corporate America. As a result the only truth in a statistic, as I've come to believe, is what you give it.

Now you may wonder what this has to do with being a DBA and hold on, I'm getting there, but indulge me a moment in looking at some statistics. What is the uptime of your servers? Recently we had to report on that and my counterpart in the engineering side of my company reported we were 99.99% available. That ought to be easy to measure, right? It's a simple calculation based on....

Well there's the rub. What do you base it on? I could easily look at the uptime as reported by the server, a simple took called uptime.exe is available on Windows systems and it reports how long the system has been running since it's boot. Now you have to keep tabs on this over multiple reboots, but if you added all these up, then theoretically you'd have the uptime. So I could look at a server that's been running for 365 days and say it's been 100% available in the past year.

Or could I? What if someone was repatching switch ports and they accidentally moved my cable on the panel and didn't catch it for 6 minutes. I've just blown a "9" and I'm only 99.99% available. Or perhaps I installed SP3 on my database. Doesn't require a reboot, but my db server is down for the time of the upgrade. Well it's not down, it's actually running in single user mode for quite a bit of the time, but for all practical purposes it's down.

Suppose nothing like that happened and I'm measuring a database query from a remote box every minute over the year and I have all that data. What if the box gets busy, say someone runs a cross join (that never happens ) and you're check doesn't return for 2 minutes. Were you down? Is your 100% blown?

I'm not trying to say we shouldn't have statistics or that they have no value, but I am trying to instill some skepticism in you so you're aware of how numbers can be manipulated. Or how they can be fudged, which might be the more important thing to be aware of in this job. After all, you might want to do some fudging. Or might need to . The main thing is that there is no "truth" in statistics. They are a bundling up of data and in doing that something is lost. That something is the detail. And that detail is what you need to be aware of even if you don't dig down into it.

So the next time someone tells you a "true" statistic, back up, hold onto your wallet, and dig up a little skepticism. Because the amount of faith you have in the numbers will be the only truth that's within reach.

Steve Jones

©dkRanch.net December 2003

Return to Steve Jones Home


You rated this post out of 5. Change rating




You rated this post out of 5. Change rating