>Thanks to all who respondied, your input was helpful.
>
>For those who were doubtful about the 100% historical uptime, I have been able to confirm that there definitely has never been a software caused outage, and the records going back 5 years show no time ever that the data was inaccesible, so it seems it i accurate. If there has been downtime, it is not readily apparent and I would not be able to press them on it.
>
>My personal opinion is that MS, Windows, and other modern IT incarnations have gotten people used to the fact that computers have glitches, and lowered our expectations. Not that consitant uptime is so common, but that it should surpise people?

I'd say most people who have been involved in computer support for a long time (in my case, over 20 years) have seen systems run continuously for a "long time". I think the best I saw was about 18 months for a Novell 3.12 server, I think an extended power outage (that outran the UPS) was what finally took it down. But it's worth noting Novell 3.x is/was not a general-purpose OS, had a simple set of functions and was not Internet-facing, so therefore not subject to attacks, and didn't need regular patches.

Twelve years with 100% availability is a whole other animal. I still can't help wondering if your client has a policy in place, where regular maintenance occurs at, say, 00:00 - 04:00 every Thursday, and that is not counted as a lack of availability. Uptime vs. availability can be defined in different ways: http://en.wikipedia.org/wiki/High_availability .

There are some availability figures at http://h20223.www2.hp.com/NonStopComputing/cache/426962-0-0-225-121.html?jumpid=reg_R1002_USEN . According to that, the best option (Tandem/HP NonStop) only reaches 5 nines, which you have already said is not good enough.

I'd say the whole project hinges on making sure all parties understand exactly what is meant by high reliability/high availability. If your client really, truly has world-leading availability as you've described, and it's important to retain it, I'd say the only way they can retain that reliability is to retain their existing back end.

If they want to add functionality, it will have to be middleware that can talk to the existing back end systems, while providing modern APIs for future use. That middleware will need to be at least as reliable as the existing back end, so again it's important to have that definition of availability nailed down.