ib1: timing out; N sends not completed

[Anthony Cornehl]

On Jun 10, 2013 6:41 AM, "Julian Stecklina" <jsteckli at os.inf.tu-dresden.de>
wrote:
>
> Hello,
>
> I have two machines connected back-to-back via Infiniband with Mellanox
> Infinihost III adapters. One machine runs Linux (Fedora 19) and the
> other 9-STABLE.
>
> I sometimes get:
>
> ib1: timing out; 47 sends not completed
> ib1: timing out; 1 sends not completed
> ib1: timing out; 56 sends not completed
>
> or similar and TCP connections will be stuck after each timeout for a
> while. It is relatively easy to reproduce this behavior with NetPIPE.
>
> Any advice?
>
> Julian
>

Hey Julian,

Just some questions to try and clarify the issue...

- which machine is the OpenSM master running on?
- what does your qkey violation count look like when you run a portinfo on
the ports?
- does the issue persist when a switch is added between the hosts?

Cheers!