Morten Torstensen wrote:
lingu wrote:
Since it is a very critical and busy server may be due to heavy
network load some hear beat signal is getting missed resulting in
shifting of service from one node to another.
For automated takeover systems, especially critical ones (tho you can
argue that any system setup with automatic takeover is critical by
definition), you should have multiple heartbeat paths. Ethernet,
serial cable, on shared disk, fibre or whatnot.
Having false takeovers due to missed heartbeat on one set of ethernet
cards could also likely be missed on another set of cards, even with a
crossover cable.
Maybe you should investigate alternate paths?
indeed, commercial cluster software like Veritas REQUIRES dual path
dedicated heartbeat networks, and highly recommends implementing storage
'fencing' so that there's no way physically possible both systems could
simultaneously mount the storage.
fencing is fairly easy with SAN storage, you instruct the SAN switch to
only allow the currently active server to have access to the storage,
and when the standby server takes over, it instructs the switch to
disable access by the old active server before enabling access by
itself. fencing with shared scsi is much harder and requires special
hardware.
_______________________________________________
CentOS mailing list
CentOS@xxxxxxxxxx
http://lists.centos.org/mailman/listinfo/centos
