MTBF - Meantime between failure
In today post I would like to talk about a software metric that can give us information related to the quality of our product and how stable it is – Meantime Between Failure.
This metric measure the time interval between the moment when a failure was fixed (and the system is stable) until the moment when a new failure is detected. We could say that this metric measure the time interval when the system is up and running.
Using this metric, we can obtain to important information related to our system:
- How stable our system is
- When the next failure will occur
I think that the second point is pretty interesting, especially when we have a system in production. Theoretically, MTBF could tell to the operational and maintenance team when next fail over will have – in this way they can be prepared for it.
MTBF can be calculated in different ways, the most simple formulate for it is the sum of all the time intervals when the system didn’t had a failure divided by the numbers of failure.
In a real product, we expect to have issues open all the time, even if we don’t like this. Because of this we need to define what kind of failures we measure (for example the severity level of them). We should measure the MTBF for critical failures (system is down, clients cannot use application anymore and so on). In a normal product, this kind of failures I would expect to be counted when we would calculate the MTBF.
I calculated the MTBF for a web application that is hosted on Windows Azure from 2011. The MTBF for this web application is around 10 months. The cause of the failures that we had until now was caused by:
- A Windows Azure Service was down and we didn't had a fail over solution for it
- Client infrastructure was down and we our web application depended on that service
In this post we saw a software metric that can be used with success when we need to predict when the next time when we’ll have a failure is. This metric can be calculated very easily and can help us to understand how stable our system is.
I invite you to calculate this metric for your own system and see what values you get.