[ofw] Setting up Infiniband over WinXp- Help Needed

Fab Tillier ftillier at windows.microsoft.com
Wed Jun 10 18:42:42 PDT 2009


Hi Ashwath,

>I am new to the world of infiniband and I am trying to set up an
>infiniband network between two Lenovo x86 Desktops (Windows Xp).

Welcome!

>Problem:-
>I am able to ping 192.168.0.2 from 192.168.0.1 however ping does not
>work the other way around i.e. from 192.168.0.2 to 192.168.0.1. I don't
>understand why this is not happening. I see that the "bind" fails but I
>dont understand why. Shouldn't it be two way? (I am using one cable to
>connect the two adaptors) Please help me. Thanks.

Check your firewall settings on the 192.168.0.1 box.  Can you access the administrative share on each node from the other (\\192.168.0.1\c$, and \\192.168.0.2\c$?)

>========================================================================
>Computer No2: 192.168.0.2
>when I ran osmtest here:
>
> C:\<mydirectory>osmtest -f -a
>Command Line Arguments
>Done with args
>        Flow = All Validations
>Using default guid 0x5ad000004e7c6
>[17:59:17:437][0388] -> osm_vendor_bind: Binding to port
>0x5ad000004e7c6.
>[17:59:17:437][0388] -> osm_vendor_bind: ERR 3B21: Unable to register
>QP0 MAD se
>rvice (IB_INSUFFICIENT_MEMORY).
>[17:59:17:437][0388] -> osmv_bind_sa: ERR 0506: Fail to bind to vendor
>SMI.
>[17:59:17:437][0388] -> osmtest_bind: ERR 0137: Unable to bind to SA

You probably have OpenSM running on this node, yes?  You can't run osmtest on the same port where OpenSM is running.

>when I ran ibdiagnet here
>
>C:\<my directory>ibdiagnet
>Loading IBIS from: C:/Program
>Files/Mellanox/MLNX_WinOF/Tools/ibdiagnet.exe/lib/
>ibis1.0
>Loading IBDM from: C:/Program
>Files/Mellanox/MLNX_WinOF/Tools/ibdiagnet.exe/lib/
>ibdm1.0
>-W- Topology file is not specified.
>    Reports regarding cluster links will use direct routes.
>-I- Using port 2 as the local port.
>-E- Fail to ibsac_bind.

Don't know the details of this tool, maybe it's running into the same problems as osmtest?  Try running OpenSM on the other node and see if the problem follows the SM or not.

-Fab



More information about the ofw mailing list