More Upgrades = More Downtime

Tom Limoncelli put up yet another nice thoughtful post on measuring the performance of a sys admin team this morning. He talks about meeting SLAs, and that greatly exceeeding an SLA is a bad thing. I appreciate his logic here. Towards the end he remarks,

“Similarly, if they want to save money you can respond with scenarios that include fewer upgrades (higher risk of security problems, less productivity due to the opportunity cost of lacking new features) or by accepting a lower SLA due to an increase in outages.”

In slightly clearer terms what he’s saying here is, “If you want to stay on the cutting edge, you are going to have more downtime.”. This is something that I think a lot of people, particularly people in management, don’t often realize. This can be mitigated by running parallel systems, but the expense of that (both hard costs and soft) is rarely justified. Of course, this makes perfect sense once it’s pointed out. Truly seamless upgrades are hard to do. In many cases, they simply aren’t possible. To the sysadmins out there, next time you need to upgrade a service, don’t forget to remind people that there’s likely going to be an avaialbility sacrifice to be made. And to the managers, remember to account for that downtime when deciding whether or not an upgraade is worth it.