[ofa-general] Re: More questions/comments on IPv6 SNM consolidation option in OpenSM

[Ira Weiny]

On Fri, 30 May 2008 05:50:24 -0700
Hal Rosenstock <hrosenstock at xsigo.com> wrote:

> Ira,
> 
> The IPv6 SNM consolidation option in OpenSM currently collapses the SNM
> groups down to 1 group aliased group. However, in a heterogeneous
> network, not all ports will be able to meet certain group parameters
> (MTU, rate). This has been discussed on the list before. My current read
> of the code indicates that these joins would be rejected. Is that
> right ?

Yes I beleive so.

>
> If so, my question is why not allow them to create and join with
> their original real multicast group for this case ?

That would be fine as long as there were not too many "odd" nodes.

>
> The downside would
> be that if there were a lot of ports like this, then the consolidation
> would reduce the number of groups but maybe not enough. So would we then
> want an additional option for doing this (and what the default should
> be) ?

I think it would be best to consolidate all the "like" ports.  For example if
you had 3 different MTU's on the fabric then you would have 3 different MGID's
and groups.  The main reason I did not do this was because it would have been a
much larger change to the code and I did not want to risk breaking things.

>
> One cut on the default would be to keep it the same as now but
> does that really matter ? Ideally, those additional SNM groups would be
> collapsed too. I think that aspect was dealt with in Jason's approach to
> this in a thread entitled "IPv6 and IPoIB scalability issue":
> http://lists.openfabrics.org/pipermail/general/2006-November/029621.html
> in which he proposed an MGID range for collapsing IPv6 SNM groups.

Ah yes...  I guess I should have read this part before responding above!  ;-)

> 
> Also, have you tried IPv6 SNM consolidation with multiple partitions ? I
> may have more on this aspect later.
> 

No, as we don't really use partitions.

Ira