[openib-general] comments on DAT registry in OpenIB
Caitlin Bestler
caitlin.bestler at gmail.com
Thu Jun 30 09:53:18 PDT 2005
We have to keep in mind that DAT was designed for a broader
purpose than what is proposed for "RDMA verbs". It was designed
to support transport *and* OS neutral applications that could
be migrated easily in a number of ways: transport, form OS to
OS, from user to kernel, etc.
So in some ways DAT is like RPC and other middleware. It
simplifies life for many applications, but it should not be the
only options available. If the RDMA verb layer is well designed
it will allow DAT (or IT-API) to be implemented simply and
at low cost. Developers can then decided whether they
prefer a higher layer API with more portability or the 'raw'
interface that is closer to the wire. With the latter they
get more control, but will have to respond to more explicit
error condiditons, etc.
That would end up looking something like:
application ---------> DAT Layer -----------> RDMA Verbs -------->
model-specific code
\ ^
\ /
\------------------------------------/
I don't think that trying to shoehorm the DAT Layer into
*being* the RDMA verbs definition makes sense, especially
if doing so involves sacrificing it's other objectives. It is,
however, an excellent interim solution that allows time
to properly specify and develop a usable verb layer
RDMA interface. And DAT will remain a valuable option
for application developers that prefer wider portability
at least one layer of abstracting of the raw results
into terms that are relevant to the application.
More information about the general
mailing list