[openib-general] Re: Problem configuring ipath_ether

[Bryan O'Sullivan]

On Sun, 2006-03-26 at 17:23 +0200, Michael S. Tsirkin wrote:

> Thanks, sounds interesting. Some questions:
> Grepping for MTU I even see 16384 and 32K - this is done with the same
> technique, isn't it?

Yes, it is.

> Any ideas on why isn't something like this done in the standard IPoIB then?

We developed ipath_ether before the kernel's IPoIB was even present (it
has worked since 2.6.0 or so), and we've been more focused recently on
the IB stack as a whole than on tweaking IPoIB.

> Anyone on the list knows whether/when this is a good idea?

There's no harm at all in at least giving it a try.  Big MTUs have been
a fact of life on "fast" networks for years, so it's not a novel
technique.  Faking big MTUs by sending a pile of packets back to back is
maybe a bit more novel :-)

> I expect it would be relatively easy to hack something like this into existing
> IPoIB codebase - I expect no one tried this yet?

Not that I know of.  It would be interesting to see results.  It might
help to offset the performance loss from the debacle over stretch
ACKs :-)

	<b

-- 
Bryan O'Sullivan <bos at pathscale.com>