[c-nsp] MTBF for Cisco products

[Tim Franklin]

On Tue, November 27, 2007 10:56 pm, rbf+cisco-nsp at panix.com wrote:

> It seems low to me.  1/4.5 = .22222
>
> So a 4.5 year MTBF would imply that about 22% of the hardware would
> fail every year.  Are you seeing a faiolure rate anywhere near that
> high with your networking equipment (from any vendor)?

I'm not sure how you get to that number without seeing the failure
distribution curve.

I'd expect routers to follow most other electronic equipment with a
bathtub curve - high infant mortality / DOA, virtually no failures through
the useful life, high failure rate again at some 'end of life' point.

If you assume infant-mortality to be '0 years', where your 'end of life'
falls depends on your infant-mortality percentage.  20% gets you 'most'
kit starts dying at 5 years, 25% at 6 years, 50% at 9 years.

Even assuming your even / flat failure distribution across the lifetime,
you need an even rate over 9 years to get a mean of 4.5, which is 11% per
year, not 22%.


Which is still not to say that it's not especially good, just that MTBF
isn't a particularly good metric by itself as to how 'bad' is likely to
hit you.

To use the car analogy, if I *know* that the fan tray^Wbelt is pretty much
guaranteed to need replacing between about 55k and 65k miles, I can plan
for that.  If all I know is that something between a bulb blowing and the
engine exploding is going to happen somewhere between 20k and 100k miles,
it's less useful.  Same MTBF, though :)

Regards,
Tim.