Re: Hi-AV cluster disk ownership?

["Jeff Darcy" <linuxguy@tambreet.com>]

> In a cluster with storage disks on SAN, what is the
> mechanism to identify which disks belong to which
> cluster?  NOTE: there can be many many nodes on the
> SAN, all of which may not be clustered with each
> other.
>
> In other words, when a node belonging to a cluster
> boots up, and scans and sees thousands of disks on the
> SAN, how will the kernel decide; which disks to
> import?

There are basically three methods in common use.  High-end disk arrays often
address the problem at the target end, by only making devices visible to
configured ports or FC WWNs.  FC-fabric gear addresses the problem in the
middle with features such as zoning, which in effect limits the portion of
the SAN that's actually "visible" from any host.  Lastly, there are all
sorts of host-based approaches.  While these operate at different levels
within the OS/BIOS/whatever, they mostly work by reading configuration
information at some "well known" location that tells the system which
devices it can access.

There are, obviously, all sorts of tradeoffs and many opinions about which
method is "best", but I won't get into that.



Linux-cluster: generic cluster infrastructure for Linux
Archive:       http://mail.nl.linux.org/linux-cluster/