[openib-general] OpenIB and OpenRDMA: Convergence on common RDMAAPIs and ULPs for Linux

[Woodruff, Robert J]

 Roland wrote, 
>I believe kDAPL-based layers are an OK short-term solution, but I
>don't think anything like this should be proposed for merging in the
>Linus kernel.  If we need another abstraction layer on top of our
>existing abstraction layer, that just says to me that we should fix
>the current abstraction layer.

> - R.

The InfiniBand core mid-layer is infiniband specific. I don't think it 
was intended to be an RDMA device independent layer. kDAPL was intended
to be that layer. 

Seems that at some level you need to have a RDMA device class specific 
layer. The IB core layer, CM, MAD services, are really pretty IB
specific
and don't apply to other RDMA types of devices. The core verbs are
similar, but the other IB mid-layer components (SA query, mad service,
CM)
are very IB specific. 

woody