[ofa-general] Re: Default multicast group rate

[Hal Rosenstock]

On Fri, 2007-04-13 at 09:58, Michael S. Tsirkin wrote:
> > Quoting Hal Rosenstock <halr at voltaire.com>:
> > Subject: Re: Default multicast group rate
> > 
> > On Fri, 2007-04-13 at 09:50, Michael S. Tsirkin wrote:
> > > > So the question is whether the best default is 2.5 Gbps which would
> > > > allow any nodes to join or whether the current default is appropriate ?
> > > > I know certain people's opinions who have been vocal on this list up to
> > > > now. I'm looking for other opinions. Thanks.
> > > 
> > > Just as a summary of what I was saying in another thread,
> > > I propose implementing a dynamic approach, where we start at
> > > 4x DDR, and drop the rate gradually as lower rate nodes join,
> > > or raise the rate when all lower rate nodes have left.
> > > 
> > > reregister can be used to let existing members know that
> > > rate has changed. Later, a spec extension can be designed
> > > to notify group members about rate change.
> > > 
> > > I think this dynamic approach would be the best default, static
> > > configurations can be supported as an option.
> > 
> > We don't have that now so I'm trying to find out whether there is any
> > consensus on which static rate should be the default currently.
> 
> It kind of seems that 1x makes sense for the HOSTS group.
> But it's a harder question for other mcast groups.

And we only get 1 choice :-( So someone is going to be unhappy whichever
way this goes. This is another one of those imperfect engineering
tradeoffs.

-- Hal

> > I also think that before a dynamic approach can be the default, we will
> > need some experience with the behavior in this mode.
> 
> Fair enough.