From: Gavin Thomas Nicol (gtn@ebt.com)
Date: Fri Jul 28 2000 - 22:58:13 PDT
> So what if a machine reboots once in a while, with clustering you
> just dont care - it means a 10% hit on performance until the machine
reboots.
>
> Comparing uptime is like comparing dick size. Nice to have, but easy to
> compensate for.
Depends on how hard the crash was, and what was going on at the time. Let's
put it this way, if you stall half way through a performance, someone might
just notice unless you're really quick with the substitute... and if you
never recover, you'll be put out to pasture.
A good example recently was when an NT server box went down. The machine
had gobs of memory, and when it crashed (TCP/IP stack corruption
apparently),
it dumped memory to disk... until the disk was full. The machine couldn't
be rebooted automatically.
The point is that you shouldn't be *required* to have a cluster. OS's should
not have hard crashes so frequently. It really annoys people when the hard
crash causes data loss, or time loss. Also, remember that clusters cost $$$$
to buy and maintain.
Again, I've seen Sun boxes crash, but not every 3 days or so.
This archive was generated by hypermail 2b29 : Fri Jul 28 2000 - 23:08:03 PDT