[openib-general] RFC: detecting duplicate MAD requests

Michael S. Tsirkin mst at mellanox.co.il
Thu Jun 15 05:43:23 PDT 2006


Quoting r. Sean Hefty <sean.hefty at intel.com>:
> Subject: RE: [openib-general] RFC: detecting duplicate MAD requests
> 
> >Well the ACK for the direction switch is special, isn't it?
> >All I'm saying, let's pass it up to the application.
> 
> I really don't think that this is the direction that we want to take the
> interface.

Yes, you are right. So, I thought about this some more, and I think
I see how your approach can be adapted without breaking applications
in subtle ways:

When a transaction arrives, pass it to user and don't keep any state.  When the
ACK for segment 0 arrives, we know there will be response in this transaction,
so we can queue it up already (but don't send yet as we don't have the data).

Start sending when user responds.

To solve the case where user responds before ACK for segment 0 arrives, a
responder in DS RMPP will pass IsDS flag when he sends the response. mad core
will then don't start sending until ACK for segment 0 arrives.

-- 
MST




More information about the general mailing list