Improving Availability vs Mean Time Between Failures

Question

Let's assume that I have to deal with an unreliable system accessed by many users that goes down/fails, let's say, X times for X minutes weekly. If only one of the following can be improved due to whatever reason, what's more important: ensuring double the availability of the service or double the mean time between failures?

Interested to hear your opinion.

Thanks

Claudio

Answer 1

If your systems falls down for T minutes X times per week, and assuming a 7/7 24/24 operation:

• its total downtime in the week is T*X.
• its relative availability A is 1-(T*X)/(7*24*60) (absolute is 7*24*60-T*X)
• its MTBF is (A * 7*24*60)/X or 7*24*60/X-T

In other words,

• increasing A means to decrease T or X or both
• increasing MTBF means to decrease X or T or both

So both are just two sides of the same coin and whatever you do to improve one will benefit the other

Edit: I didn't notice the "double" in your question:

I'll assume that T is rather constant (e.g. time to reboot the server):

• You can't double availability if its above 50%. But if availability is lower, doubling it means achieving a very high improvement.

• You can always double the MTBF. All it takes is approximately halving X.

Answer 2

[This started as a comment, but I ran out of room. So, I'm posting as an answer.]

If only one of the following can be improved due to whatever reason, what's more important: ensuring double the availability of the service or double the mean time between failures?

The relative importance of availability vs. MTBF depends on the usage of the system. I'm going to assume that user's work is lost when failures occur.

If the users work in short sessions (much shorter than MTBF), then work is rarely lost to failures. In this case, the uses might not even notice failures. In this case, availability may be a greater issue than MTBF.

If the users work in long sessions (comparable in length to MTBF), then work is more likely to be lost due to failures. Loss of work tends to be more annoying than unavailability of the system. In this case, MTBF may be a greater issue than availability.