[ofa-general] Re: [PATCH 1/4] ib/ipoib: specify TClassand FlowLabelwith PR queries for QoS support

[Jason Gunthorpe]

On Wed, Aug 08, 2007 at 01:31:04PM +0300, Michael S. Tsirkin wrote:
> > AFAIK, when using the 64K MTU setting for IPoIB, if the remote side
> > doesn't support RC then things go wonky.
> 
> The scenario you describe works fine for me.
> What does go wonky for you?

Only this path MTU buisness, which is somewhat unusual. PMTU is
difficult to deal with in UDP applications if you want to prevent
fragmentation. Wonky is maybe not a good term.

> > For TCP things *might* be saved
> > by path mtu discovery -
> 
> MTU discovery is not restricted to TCP AFAIK.

That is something I admit I've never looked at in Linux, if PMTU
affects the route cache or the socket.. 

> > but PMTU is driven by ICMP errors which are not generated by an IB
> > network.

> IPoIB CM code generates them in software.

Oh! That solves it for RC then. Does this happen for UD as well?

Jason