I recently ran into a customer asking about availability for a solution we were proposing for them. They were asking all kinds of questions, like mean time before failure(MTBF), availability - especially in a clustered environment, etc. So, I did a fair bit of research on the subject and came up with the following.
Availability is the number that people look for in a solution. How many 9's is a key number when looking at a solution. In digging around, I found that availability is based on two factors; one, MTBF, which is published by the manufacturer, and two, mean time to repair(MTTR), which is based upon a number of factors, like how quick can you get replacement hardware and how quick can you get the replacement hardware configured properly.
Once you have MTBF and MTTR, availability is calculated as MTBF / (MTBR + MTTR). So, if you have a piece of equipment that is has a MTBF of 51,000 hours and you estimate your MTTR at 48 hours, your availability is 99.9%, or 8 hours of downtime per year. Not a great availability number, but something some people can live with.
If you want to increase your availability, you should consider redundancy, like active/passive, where you have one node that is operating and if it fails, the second node takes over. If you have two nodes with an availability of 99.9%, your availability becomes 99.9999%, or under one minute of downtime per year.
How did I get that? The answer is all statistics. A box with an availability of 99.9% has a 0.1% change of failure. If I have two boxes with 0.1% chance of failure, then the chances of both boxes failing at the same time is 0.1% * 0.1% or 0.0001%. Thus, your availability becomes (1 - 0.0001%) or 99.9999%.
Sounds good right, well if you want even better uptime, try three boxes and your availability becomes 99.9999999%, or 0.03 seconds of downtime per year. Now that is impressive!
Comments
Post new comment