Hi,
On Mon, 6 Dec 1999 07:35:25 -0600 (CST), louieorbeta@technologist.com
(Louie Robles Orbeta) said:
>> What happens if your network connection dies but the scsi is still live?
>> Both machines start talking to the disk. It's dead, Jim.
> I think another hurdle to overcome is to find a way to release the
> SCSI reserves in the event of a failover. That way, the fallback node
> can mount the SCSI drives R/W...
SCSI has ways to do that just fine. The real problem is how to make
100% sure that the node which had its reservation stolen cannot continue
to write to the disk. XFS solves this by hardwiring a control line to
the other node's power line to kill it dead on failover. :)
With Fibrechannel fabrics, things are much better: you can tell the
fabric to fence off particular nodes, and the disk will then reject IOs
from it automatically. FC makes a lot of this stuff much easier.
--Stephen
On Mon, 6 Dec 1999 07:35:25 -0600 (CST), louieorbeta@technologist.com
(Louie Robles Orbeta) said:
>> What happens if your network connection dies but the scsi is still live?
>> Both machines start talking to the disk. It's dead, Jim.
> I think another hurdle to overcome is to find a way to release the
> SCSI reserves in the event of a failover. That way, the fallback node
> can mount the SCSI drives R/W...
SCSI has ways to do that just fine. The real problem is how to make
100% sure that the node which had its reservation stolen cannot continue
to write to the disk. XFS solves this by hardwiring a control line to
the other node's power line to kill it dead on failover. :)
With Fibrechannel fabrics, things are much better: you can tell the
fabric to fence off particular nodes, and the disk will then reject IOs
from it automatically. FC makes a lot of this stuff much easier.
--Stephen