Re: replicating, balanced web-server with *write* access?
<quote who="Christian Hammers">
> On Sun, Nov 11, 2001 at 02:09:01PM +1100, Jeff Waugh wrote:
> > RAID on Network Block Devices. You get the benefits of RAID, but over a
> > number of different machines, perhaps even on different networks if the
> > topology allows for the performance requirements.
> Does it really allow writing in *both* directions? I mean both servers
> should be able to write to the same "filesystem" so they would have to
> mount each other as nbd... Else it would only be good for one-way failover
Okay, how about this... It's ASCII ART TIME!
_____ _____ _____
| | | | | |
| ND1 | | ND2 | | ND3 | NBD device machines: 1, 2 & 3
|_____| |_____| |_____|
| | | |
| FS1 | | FS2 | File server machines: 1 & 2
| CL1 | Client machine: 1, for the sake of the image. :)
The RAID member machines all run an NBD server, so let's say we have three
network devices to make our RAID with. The two fileservers are for failover,
so we really only use one. It uses the NBD devices, and operates the RAID.
Our client machine uses the filesystem on the fileserver (however it needs
it, it could be samba, nfs, appletalk, etc).
If an NBD device machine goes down, the fileserver handles this as it would
any other RAID situation. When the machine comes back up, the NBD can be
resynced with the others.
If a fileserver machine goes down, bring up the other one on the same IP
address with heartbeat. It can also bring up the NBD devices and get the
RAID going again.
If the client goes down, thwack them on the head. ;)
> > It's A CRAZY SCHEME, but it MIGHT JUST WORK! 
> yeah, that's what I want to have on my production servers <g>
That's the spirit! LINUX UBER ALLES! ;)
"Basically my philosophy on release management is that it should be
like police brutality." - Maciej Stachowiak