On Sun, Nov 09, 2008 at 10:25:29AM -0600, lee wrote: > On Sun, 9 Nov 2008 08:50:28 -0500 > "Douglas A. Tutty" <[EMAIL PROTECTED]> wrote: > > > Parts break, > > redundancy kicks in, change the dead part, still the same computer. > > If so, you can do that with three cheap i386 boxes. > > Let's say you have a router/firewall/proxy, a fileserver, a mailserver > and a webserver. How do you make it so that each of these has a backup > server that automatically and seamlessly kicks in when needed?
Each "box" is two boxes, running OpenBSD with CARP :) Of course, they're all the exact same hardware and you have a backup tape for each type of box. When one box dies, you "restore" it to an off-the-shelf replacement and get its mate redundant again. Then you fix the broken box. That's where I got the "three" from: two hot-redundant (i.e. slowly wearing out) and one cold-spare (slowly ageing out). In the scenario with multiple pairs of hot-redundant, I've suggested that they could share the cold-spare. At some point of scale, you go with something with LPARs and have fewer overall boxes and higher reliability. Where that point is, I don't know. I'll never have the money even for something like the IBM p5-570 (up to 16 processors, up to 64 micro-partitions). The last time I looked, the shared cost of a micro-partition was about $400 each but gave you the performance of a $1000 server. With LPARs, you don't have to dedicate 2 to redundancy, since the hardware/firmware takes care of that at a lower level. Of course, their $400 per partition/server likely doesn't cover the service contract. :) Doug. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
