[openib-general] CM MRA handling

[Fab Tillier]

> From: Sean Hefty [mailto:sean.hefty at intel.com]
> Sent: Tuesday, February 01, 2005 9:06 AM
> 
> >I was thinking about how to handle receipt of the MRA CM message.  Rather
> >than add timer logic to the CM code base, why not leverage the timeout
> >facility of the MAD agent?  When receiving an MRA, cancel whatever MAD is
> >in progress (REQ, REP, or LAP), and reissue the exact same MAD with a
> >timeout as requested in the MRA.  This allows using the same MAD timeout
> >logic for MRA_XXX_RCVD as for the original message.
> >
> >The down side of this is that you end up issuing an extra message for
> every
> >MRA received (in effect responding to each MRA) with a repeat of the
> >message that was MRA'd).
> 
> I thought about doing this exact same thing.  But I didn't like the idea
> of
> sending a response to the MRA, which could generate a second MRA.  Also,
> I'm
> guessing that handling timewait will require adding timer logic to the CM
> code base anyway, so I'm hoping to leverage that.
> 

Looking through the spec some more, it's pretty clear that the CM should not
resend the message for which the MRA was received until the indicated time
in the MRA plus the packet lifetime for the path.

- Fab