On Wed, 2008-04-09 at 19:56 +0200, Bernd Schubert wrote: > Hello Yevgeny! > > On Tuesday 08 April 2008 22:22:38 Yevgeny Kliteynik wrote: > > Sasha Copyist wrote: > > > Hi Bernd, > > > > > > [adding Yevgeny..] > > > > > > On 11:35 Tue 08 Apr , Bernd Schubert wrote: > > >> On Tuesday 08 April 2008 03:44:06 Sasha Copyist wrote: > > >>> Hi Bernd, > > >>> > > >>> On 11:47 Fri 04 Apr , Bernd Schubert wrote: > > >>>> opensm-3.2.1 logs some error messages like this: > > >>>> > > >>>> Apr 04 00:00:08 325114 [4580A960] 0x01 -> > > >>>> __osm_state_mgr_light_sweep_start: ERR 0108: Unknown remote side for > > >>>> node 0 > > >>>> x000b8cffff002ba2(SW_pfs1_leaf4) port 13. Adding to light sweep > > >>>> sampling list Apr 04 00:00:08 325126 [4580A960] 0x01 -> Directed Path > > >>>> Dump of 3 hop path: Path = 0,1,14,13 > > >>>> > > >>>> > > >>>> From ibnetdiscover output I see port13 of this switch is a > > >>>> switch-interconnect (sorry, I don't know what the correct > > >>>> name/identifier for switches within switches): > > >>>> > > >>>> [13] "S-000b8cffff002bfa"[13] # "SW_pfs1_inter7" lid > > >>>> 263 4xSDR > > >>> > > >>> It is possible that port was DOWN during first subnet discovery. > > >>> Finally everything should be initialized after those messages. Isn't it > > >>> the case here? > > >> > > >> I think everything is initialized, but I don't think the port was down > > >> during first subnet discovery, since the port is on a spine board (I > > >> called it 'inter') to another switch system. We also never added any > > >> leafes to the switches. > > > > > > It is interesting phenomena then. > > > > > > Yevgeny, do you aware about such issue with Flextrinocs switches? > > > > I've seen it before. It means that during discovery some switch has > > answered NodeInfo query, but then when OpenSM started to query for > > PortInfo for each port of this switch, switch didn't answer for some > > (or all) ports. > > > > I think that this might happen if a switch has just been "plugged in", > > and internal switches are doing autonegotiation - they are bringing > > ports up and down when determining whether a link is SDR or DDR. > > > > In any case, this "phenomena" should disappear after a couple of > > dozens of seconds, when all the autonegotiation phase would be over. > > > > Bernd, am I close? > > > > We never plugged in additional switches and the message appear on each opensm > startup. However, the messages appear only once after opensm was started, but > then never again. Would the switches do a SDR/DDR negotiation on opensm > startup?
Links perform physical negotiation independent of SM. > And since we are at SDR/DDR, it also might be related. Hal and I are also > discussing an odd SDR/DDR ibnetdiscover problem. Ibnetdiscover just thinks > some ports are at SDR, while ibstatus and perfquery do tell these ports are > at DDR. I'm not sure the link speed is "stable". -- Hal > Thanks, > Bernd > > _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
