[PATCH] opensm: Special Case the IPv6 Solicited Node Multicast address to use a single Mcast (WAS: Re: [ofa-general] IPoIB, OFED 1.2.5, and multicast groups.)

[Ira Weiny]

On Tue, 15 Jan 2008 19:33:47 +0000
Sasha Khapyorsky <sashak at voltaire.com> wrote:

> On 11:17 Tue 15 Jan     , Hal Rosenstock wrote:
> > On Tue, 2008-01-15 at 19:20 +0000, Sasha Khapyorsky wrote:
> > > On 09:37 Tue 15 Jan     , Hal Rosenstock wrote:
> > > > On Tue, 2008-01-15 at 08:47 -0800, Ira Weiny wrote:
> > > > > On Mon, 14 Jan 2008 17:40:33 -0800
> > > > > Hal Rosenstock <hrosenstock at xsigo.com> wrote:
> > > > > 
> > > > > > On Tue, 2008-01-15 at 01:43 +0000, Sasha Khapyorsky wrote:
> > > > > > > On 17:22 Mon 14 Jan     , Hal Rosenstock wrote:
> > > > > > > > On Tue, 2008-01-15 at 00:50 +0000, Sasha Khapyorsky wrote:
> > > > > > > > > On 16:05 Mon 14 Jan     , Hal Rosenstock wrote:
> > > > > > > > > > On Mon, 2008-01-14 at 15:35 -0800, Ira Weiny wrote:
> > > > > > > > > > > On Mon, 14 Jan 2008 12:23:34 -0800
> > > > > > > > > > > Hal Rosenstock <hrosenstock at xsigo.com> wrote:
> > > > > > > > > > > 
> > > > > > > > > > > > On Mon, 2008-01-14 at 10:51 -0800, Ira Weiny wrote:
> > > > > > > > > > > > > Hey Hal, thanks for the response.  Comments below.
> > > > > > > > > > > > > 
> > > > > > > > > > > > > On Mon, 14 Jan 2008 12:57:45 -0500
> > > > > > > > > > > > > "Hal Rosenstock" <hal.rosenstock at gmail.com> wrote:
> > > > > > > > > > > > > 
> > > > > 
> > > > > <snip>

<snip>

> > > > > 
> > > > > > 
> > > > > > No, MLID is not the group (at least in IBA terms); I was referring to
> > > > > > the base SNM MGID (with partition and low 24 bits masked off).
> > > > > 
> > > > > You are right the MLID is not the group.  But the patch only creates 1 MGID as
> > > > > well. 
> > > > 
> > > > I think you are confusing internal implementation with the outside view
> > > > of what is going on.
> > > > 
> > > > >  I think I see where you are coming from but...
> > > > > 
> > > > > Let me ask this.  (I think I know the answer but I will ask anyway.)  If you
> > > > > have 3 MGID's (0xFF...1, 0xFF...2, 0xFF...3, 3 actual mgrp structures in
> > > > > opensm) and you map them all to MLID 0xC001 will a message to 0xC001 reach all
> > > > > 3 nodes?
> > > > 
> > > > Ignoring the partitioning (and assuming the rates and MTUs are all the
> > > > same), then yes.
> > > > 
> > > > Let me ask you this:
> > > > Do all IPv6 SNM MGIDs show up when you do SA GetTable for groups ?
> > > 
> > > Let me answer. No. Only first MGID will be shown.
> > > 
> > > I guess it is wrong in terms of IBTA,
> > 
> > Yup; all those MGIDs need to be able to be queried.

I _will_ take your word for this, but I am still curious as to who is going
to know these MGIDs to be queried?

> > 
> > > but since whole feature is optional
> > > I don't think it is a disaster (rather known limitation), and ipv6 not
> > > working with big clusters is.
> > 
> > Yes, but IMO we can/should do better than this.
> 
> Nobody said anything against it.

Totally agreed.

Ira

> 
> > It's not much more work,
> > is it ?
> 
> I guess it is more work, don't know for sure yet.
> 
> Sasha
> 

<snip>