[dm-devel] RE: How is the multipath pg policy group-by-nodename any differen t than the failover policy?

[Lars Marowsky-Bree]

On 2005-05-03T13:22:32, Arthur Bergman <sky at nanisky.com> wrote:

> >How is the resultant path grouping any different than if you used the
> >failover
> >path grouping policy instead?
> >
> >--
> >dm-devel mailing list
> >dm-devel at redhat.com
> >https://www.redhat.com/mailman/listinfo/dm-devel
> >
> 
> Huh, as far as I understand it group_by_nodename uses the
> 
> /sys/class/fc_transport/target0:0:0/node_name information
> 
> from our setup, there are 8 paths to two controllers, there is no 
> failover cost but it defeats read caching pretty beadly to use both 
> controllers..
> 
> /sys/class/fc_transport/target0:0:0/node_name:0x50050768010003d5
> /sys/class/fc_transport/target0:0:1/node_name:0x5005076801000361
> /sys/class/fc_transport/target0:0:2/node_name:0x50050768010003d5
> /sys/class/fc_transport/target0:0:3/node_name:0x5005076801000361
> /sys/class/fc_transport/target1:0:0/node_name:0x5005076801000361
> /sys/class/fc_transport/target1:0:1/node_name:0x5005076801000361
> /sys/class/fc_transport/target1:0:2/node_name:0x50050768010003d5
> /sys/class/fc_transport/target1:0:3/node_name:0x50050768010003d5
> 
> 
> So all path that ends at a matching node_name, even if they are on 
> different transport targets get one path.

Now figure in that this might need to be stacked with more hw-specific
stuff like dm-emc ;-)

I think that ought to work already if multipath-tools would set up the
right cascaded multipath tables in theory though; however it might not
because depending on how the stacking works some things might not
propagate up or down correctly right now.


Sincerely,
    Lars Marowsky-Brée <lmb at suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"