iSCSI support

[Richard Burakowski]

oxo at rucus.net wrote:

>I have 3 datacentres connected by 12 core gig fibre (only using one pair
>at the moment, but the fibre is there for future use) each connected
>directly to the others.  I want a system that I can start off with one
>disk server in one datacentre, and then step it up to have mirrored disk
>servers in each of the other datacentre's which are kept up to date in
>real time and can take over instantaneously if one of the others fails.
>
>It must also be scalable (non destructive resizing of the system) and
>support both linux and FreeBSD.  I am willing to wait for this, but can
>anyone point me in the right direction.  iSCSI seems to be it, but I'm
>not sure.
>  
>
all, don't get network attached storage confused with network attached 
filesystem confused with clustered filesystem.

if you go for fibre channel network attached storage, it dosen't matter 
if the host and storage array are in the same cabinet, across the room 
or in different data centers.  if your requirement is only to have one 
host up at any time then it can raid1 3way mirror over the sites.

of course it gets really messy when one of the links goes down and you 
have to decide if it really has and not just the way your testing, who 
becomes master and enforce it so there's no corruption (if the "down" 
host continues writing).

you mention multiple cores and the datacenters connectected in a ring, 
which means you can multipath in both directions of the loop.  don't 
know of any fc multipathing for freebsd.

doing this in iscsi will be a lot cheaper.  switches will be gigE with 
fibre uplinks to connect the sites.  targets and initiators can be 
regular boxes with more/less/none directly attached disks, all connected 
via gig nics.  multipathing/link failures are handled by routing 
daemons/protocols which already exist.