[ofa-general] [Bug 450] New: IPoIB BW drop (measured with iperf) with mtu=1500 on x86 RH4UP3

[Moni Levy]

On 3/14/07, Michael S. Tsirkin <mst at dev.mellanox.co.il> wrote:
> > Quoting Moni Levy <monil at voltaire.com>:
> > Subject: Re: [ofa-general] [Bug 450] New: IPoIB BW drop (measured with iperf) with mtu=1500 on x86 RH4UP3
> >
> > On 3/13/07, Michael S. Tsirkin <mst at dev.mellanox.co.il> wrote:
> > >> >Why are you playing with ifc mtu?
> > >>
> > >> We needed to do some IP forwarding tests between IB and Ethernet and
> > >> wanted to have the same MTU for the eth0 and ib0 interfaces, after
> > >> that we reproduced that in a peer to peer configuration.
> > >
> > >OK but PMTU discovery would do this better, won't it?
> >
> > I'm not sure (it will probably drop some packets at the beginning).
>
> Can you check pls?
>
> > Anyway lower MTU should work.
>
> Manually tweaking MTU seems to be broken in lots of systems -
> I sometimes see the same behaviour with gigabit ethernet.
>
> > >As a side note, with 1.5K MTU it's probably better to use datagram mode
> > >anyway.
> >
> > Now I see that I missed that in the report. We used datagram mode.
>
> Did you try checking ethernet BW on the same machine?

Ethernet seem to work fine.

I can't reproduce the bug in the last week. The only thing that I see
now is that at that same MTU and connected mode enabled I get the same
low BW. What version of tcpdump are You using ?

-- Moni

>
> --
> MST
>