[ofa-general] ***SPAM*** what are inmpications of opensmd going down

[Ira Weiny]

On Tue, 5 Aug 2008 11:06:39 -0400
"Hal Rosenstock" <hal.rosenstock at gmail.com> wrote:

> Hi,
> 
> On Tue, Aug 5, 2008 at 9:58 AM,  <kovlensky at interia.pl> wrote:
> > Hi all,
> >
> > I\'m trying to find any informations on opensmd disappearing from the network.
> 
> In general, SM should always be present.
> 
> > Let\'s assume we\'ve got fully running and operational network, with lids assigned. My undestanding is that opensmd does nothing then and hca\'s with lids assigned can happily survive without it?
> 
> Yes and no. SA requests (like path requests and multicast) and subnet
> changes will not be serviced during that time.
> 
> > In other words - having opensmd going down will stop new nodes from joining the network (ports will stay in init state), but will not disrupt the ones which were up?
> 
> Yes.
> 
> > Also opensmd, when configured to not reassign lids, will not break it when going back up, as it will preserve current setup. Right?
> 
> Yes, as long as no new nodes are present which cause a lid conflict
> (e.g. subnet merge when SM was offline).
> 
> > Conclusion - taking opensmd down and up should not influence ib network in any way with exception of new hca\'s appearing.
> 
> Also switches or links which fail when SM is offline.
> 
> -- Hal
> 
> > All above comes from my logic and, as I cannot find any data to confirm that, could anybody confirm that or point at any error there? I\'m considering migrating to opensm and this information is vital to me.
> >

I agree with everything Hal said.  But I am going to assume that you only want
to be able to restart OpenSM rather than actually run without it?  If so I can
tell you that we have done that often and have not seen any detriment to
performance on a "normal"[*] basis.  ([*] Obviously there might be bugs and
corner cases.)

If you are proposing to run without an SM for an extended time, that will
probably cause a problem.

Ira