Collections of (physically) small computers

Adrian Graham wrote: Although working out the cost of those HP servers in terms of power, reliability, security, systems management and user time wasted due to bad software - then rinsing and repeating every couple of years due to the upgrade cycle - might make for some interesting numbers. But as I said in a message to the Bletchley list just now, if we all wanted a centralised, thin-client way of working, we'd all be doing it already because the technology has never gone away. OK, so that's typically based around a handful of core servers rather than a single giant mainframe, but from a user and managerial perspective it looks the same - but hasn't been adopted in wide numbers. Aside: Anyone here have experience of mainframe design? I mean, sure they're reliable and fast and everything - but how much of the cost is due to design effort and how much is purely due to the fact that they don't sell many and need to pay the bills? (I just had this vision of an IBM engineer seeing how many buses, CPUs and I/O interfaces they could draw on a napkin, then adding in a few redundant power supplies and calling it done ;-) cheers Jules