> 1.  If I am not using rdma_join_multicast, would I simply use multiple
> calls to ibv_attach_multicast?

Yes
 
> 2.  My IB code is serving as a pass through layer between two
> pre-existing simulations which previously communicated via reflective
> memory.  It is much easier for me to insert this layer rather than
> modify these existing apps.  So SimApp A passes data to my IBLayer
> which multicasts it to IBLayers on other machines.  Instances of these
> IBLayers, upon receiving hte data, place it in shared memory for
> SimApp B to consume.  There are multiple types of data passing between
> the SimApps and my IBlayer...  All I want my IBLayer to do is take N
> Bytes of Data and place it in the correct shared memory location.  But
> if it doesn't know what type of data it is, it won't know the correct
> shared memory location...  I thought if I have multiple multicast
> groups and I use each one for a different type of data, e.g. MG 1 for
> DataType 1, MG2 for DataType 2 and so on.  Then the IBLayers receving
> the data would know that data received from MG 1 goes in shared mem
> segment 1 etc.

Work completions contain the source QP number and source LID.  I don't think 
you can tell which multicast group specific data was sent on.  You would 
probably need to carry that information in a header.  You could try using 
immediate data for that.  Btw - multicast traffic is unreliable, so you would 
need to implement your own ack / retry mechanism if you wanted reliability.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to