[Linux-cluster] wishing to run GFS on iSCSI with redundancy

Lon Hohberger lhh at redhat.com
Fri Jul 6 16:31:19 UTC 2007


On Fri, Jun 29, 2007 at 10:49:35AM +1200, Rohit Grover wrote:
> Ok, couple of notes:
> >
> >* MD is only unsafe in a cluster if it's used on multiple cluster nodes.
> >That is, it should be fairly easy to implement a resource agent which
> >assembles MD devices from network block devices - on one node at a time.
> 
> True. I would like to have MD assembling iSCSI initiators (the same set, of
> course) at multiple nodes. This will facilitate load distribution.
> Isn't it true that if MD is made to not cache any data flowing through it
> (and leave GFS to do caching and coherency control across the cluster), then
> MD should be a viable solution to putting together iSCSI initiators with
> RAID?
> 
> 
> * DRBD only will work with two writers (if 0.8.x+).  I'm not sure how many
> >mirror targets you can maintain.
> 
> 
> Could you please elaborate on this? I don't understand what is meant by
> 'DRBD will only work with two writers'. Thanks.

I think drbd only supports two-node mirrors.  I.e., you can set up a
mirror from only two nodes at a time, and that's it: no other nodes can
access the data (directly, at least).

I'm not sure how many sets of these mirror setups you can run on a given
pair of nodes, or if there is even a maximum.

> * Aren't most iSCSI targets RAID arrays already (?)
>
> Yes, they are in our case. But we also want to survive software/firmware
> failures of the iSCSI targets.
> 

Complexity has its own problems :)


-- 
Lon Hohberger - Software Engineer - Red Hat, Inc.




More information about the Linux-cluster mailing list