Is it true that the main and backup servers have to be on the same
subnet - XXX.YYY.ZZZ.* ?
Yes, if the backup server should take over seamlessly on crash of the
primary server. The reason are the network connections: the takeover by
the backup server is transparent to the VM and for keeping open the
running network connections, the VM must be accessable by the same ip
address on the target network. This is not the case if the backup server
is in a different subnet. As far as I know there are no mechanisms like
e.g. mobile ip for enabling a migration into a different subnet (yet).
The main problem I see for using different subnets (and hence the
servers beeing connected via several routers/switches) is the
performance degradation because of the higher network latency.
Or even more restricted than that, that they
have to be connected by hardwire somehow, with no intermediate switches?
No, a switched network can be used without problems although a separate
direct connection between the two servers can increase the performance.
Our primary reason for thinking of Remus is catastrophic fail-over for
important servers like license servers and such, which are accessed
globally. If the main and backup servers have to be in the same
building and it gets hit by a tornado which blows them both away Remus
isn't going to help much.
Yes, that's true, but perhaps for these kinds of failures remus is not
the right approach... a regular copy onto a dedicated backup server
would be better suited I think. Remus is designed for high availability
setups where it is important that the downtime due to a server crash is
minimal and all VM state is preserved.
Does anyone have Remus working on SuSE 10.1? If so, what kernel did
I have a running installation with kernel 2.6.31 (both dom0 and domU)
and xen 4.0.0 on arch linux, not tried on suse yet :)
Xen-users mailing list