***SPAM*** Re: [ofa-general] link width problem of Qlogic 9024 unmanaged switch
Ira Weiny
weiny2 at llnl.gov
Tue Apr 21 11:27:17 PDT 2009
Hi Yicheng,
On Fri, 17 Apr 2009 15:07:54 -0500
Yicheng Jia <YJia at tmriusa.com> wrote:
> Hi Ira,
>
> >Yea that is going to be a problem. The problem is that effectively you
> just
> > disabled the connection to the switch. A reset disables then enables
> the
> > port. Once the port is disabled the command can't talk to the switch
> any
> > longer. You will have to either reset the switch (power cycle) or go to
> > another node and enable the port. From the output you sent me it looks
> like
> > you don't have any other nodes on the switch, so I take it you are
> resetting
> > the switch to get the link to come back?
>
> Thanks your explanation now everything is clear. Can I do "reset" by
> down/enable instead of down/disable/enable so that I can reset the peer
> port on the switch?
Not that I know of.
>
> > I thought there was a warning in the man page or in the help regarding
> this
> > situation but I don't see it now.
>
> There's a warning if I try to reset a port which is not on the switch.
>
> > BTW, What are you trying to achieve with this command?
>
> Sometime there's 1x link on the subnet after reboot our system, which
> consists of several HCA nodes directly connected with the switch. By
> reboot, I mean restart each node. I am trying to achieve 4x link width by
> using this command on 1x link port. Do you have any better idea of
> resolving this problem?
The way we do it around here is to go to another node and issue the request.
I have a perl script in my "pragmatic infiniband utilities" (ibbouncelinks.pl)
which will skip the port it is running on. That tarball can be found here:
https://computing.llnl.gov/linux/piu.html
I did not code an option to look for 1X links but I think it would be simple
to do so.
Thanks,
Ira
>
> Thanks!
> Yicheng Jia
>
>
>
>
>
> Ira Weiny <weiny2 at llnl.gov>
> 04/17/2009 12:45 PM
>
> To
> Yicheng Jia <YJia at tmriusa.com>
> cc
> general at lists.openfabrics.org, Hal Rosenstock <hal.rosenstock at gmail.com>
> Subject
> Re: ***SPAM*** Re: [ofa-general] link width problem of Qlogic 9024
> unmanaged switch
>
>
>
>
>
>
> On Fri, 17 Apr 2009 12:06:43 -0500
> Yicheng Jia <YJia at tmriusa.com> wrote:
>
> > Hi Ira,
> >
> > Here is the output of "iblinkinfo.pl -R":
> >
> > ++++++++++++++++++++++++++++++++++++++++++++++++
> > [root at ib_manager ~]# iblinkinfo.pl -R
> > Switch 0x00066a00d90009c1 InfiniCon System InfinIO 9024 Lite:
> > 7 1[ ] ==( 4X 2.5 Gbps Active / LinkUp)==> 6 1[ ]
> "MT2520 4 InfiniHostLx Mellanox Technologies" ( )
> > 2[ ] ==( 4X 2.5 Gbps Down / Polling)==> [ ]
> "" ( )
> > 3[ ] ==( 4X 2.5 Gbps Down / Polling)==> [ ]
> "" ( )
> > 4[ ] ==( 4X 2.5 Gbps Down / Polling)==> [ ]
> "" ( )
>
> [snip]
>
> >
> > And the "ibstat" output:
> > +++++++++++++++++++++++++++++++++++
> > [root at ib_manager ~]# ibstat
> > CA 'mthca0'
> > CA type: MT25204
> > Number of ports: 1
> > Firmware version: 1.2.0
> > Hardware version: a0
> > Node GUID: 0x0002c90200230784
> > System image GUID: 0x0002c90200230787
> > Port 1:
> > State: Active
> > Physical state: LinkUp
> > Rate: 10
> > Base lid: 6
> > LMC: 0
> > SM lid: 6
> > Capability mask: 0x02500a6a
> > Port GUID: 0x0002c90200230785
> > ++++++++++++++++++++++++++++++++++++++++++++++
> >
> > The reset command I am using is "ibportstate 7 1 reset", I also tried
> > "ibportstate -D 0,1 1 reset", and it fails with the same result.
> >
>
> Yea that is going to be a problem. The problem is that effectively you
> just
> disabled the connection to the switch. A reset disables then enables the
> port. Once the port is disabled the command can't talk to the switch any
> longer. You will have to either reset the switch (power cycle) or go to
> another node and enable the port. From the output you sent me it looks
> like
> you don't have any other nodes on the switch, so I take it you are
> resetting
> the switch to get the link to come back?
>
> I thought there was a warning in the man page or in the help regarding
> this
> situation but I don't see it now.
>
> Also, this becomes worse if you disable the port the SM is on. (Which I
> see
> you are doing.) So you will have a noticeable delay while the SM rescans
> the
> network which it is now seeing "again" for the first time.
>
> BTW, What are you trying to achieve with this command?
>
> Hope this helps,
> Ira
>
>
> _____________________________________________________________________________
> Scanned by IBM Email Security Management Services powered by MessageLabs.
> For more information please visit http:// www. ers.ibm.com
> _____________________________________________________________________________
>
>
>
> _____________________________________________________________________________
> Scanned by IBM Email Security Management Services powered by MessageLabs. For more information please visit http:// www. ers.ibm.com
> _____________________________________________________________________________
--
Ira Weiny
Math Programmer/Computer Scientist
Lawrence Livermore National Lab
weiny2 at llnl.gov
More information about the general
mailing list