[ofa-general] Re: multicast join failed for...

[Sasha Khapyorsky]

On 16:57 Fri 13 Apr     , Michael S. Tsirkin wrote:
> > Quoting Hal Rosenstock <halr at voltaire.com>:
> > Subject: Re: [ofa-general] Re: multicast join failed for...
> > 
> > On Fri, 2007-04-13 at 09:38, Michael S. Tsirkin wrote:
> > > > > If the group is created at a lower rate, there would be no problem.
> > > > > But the default configuration should be "plug an play".
> > > > 
> > > > So you are arguing for 1x SDR as the default. We've discussed and
> > > > disagreed on this before as I think it masks performance issues and
> > > > those are harder to find. I could be wrong about this.
> > > 
> > > No, I'm arguing for dynamic configuration as the default.
> > > so we start at 4x DDR and bbring the rate down as slower nodes join.
> > 
> > OK that answers a different question I was wondering about. Or speed it
> > up if all nodes are say 4x DDR.
> > 
> > What I was trying to say was that since we don't have dynamic rate
> > support now (and I'm not signing up to do this, is someone ?),
> 
> I don't know too much about opensm yet, but I can try looking into it,
> or try talking someone into this :)
> 
> But I'm happy we all agree it's a good idea.

I stated in another thread - I don't like idea about full clients
reregistration.

Isn't it simpler to just specify desired rate and mtu for ipoib multicast
groups in config or/and as command line option? This should be simple:
if one wants to use slow ports - this will work as needed from beginning,
if not - those ports will not be able to join.

> Let's add this to osm/doc/todo?
> 
> > I was
> > saying that a static rate default of 1x SDR would eliminate the join
> > errors (at the debug "expense" of what I think are harder to find
> > performance issues). Sorry I didn't make that clear before.
> 
> I think if we either
> 1. Add an option to disable 1x support at endnode
> or
> 2. Implement a tool to find and report 1x links

It is good thing to do, but it is not directly related to the problem.
OpenSM should be reasonable regardless to how and when external tools
were used.

> or
> 3. By default, report 1x links in opensm log as errors

This also could be useful, but again I don't think it is related
directly to the problem of mixed subnets.

Sasha