Re: TCP RDMA option to accelerate NFS, CIFS, SCSI, etc.

[Kacheong Poon <Kacheong.Poon@eng.sun.com>]

> As an aside I think the RDMA proposal has a lot of holes too.  For
> example, there are in-kernel HTTP accelerators that do the complete
> client header parse and initial packet response in the hw interrupt
> handler.  There are no user buffers involved, and static response
> data is DMA'd directly from the filesystem page cache.

Even without the above special acceleration, section 2 in the draft
overestimates the number of copies involved in most OSes nowadays.

> In the case of a server response, RDMA benefits the client, not the
> server, so I fail to see why your example is problematic.  Zero copy
> send is not what this standard addresses.

Even without RDMA, people have done zero copy receive with TCP/IP before. 
What I see in RDMA is a way to provide message boundary in TCP so that apps
do not need to spend time on that (I call them lazy apps). I don't see right
away the other claims in the draft.  It seems to me that the authors should
re-evaluate their claims and provide support to them.  I don't see any
arguments in the draft.

If message boundary is the only benefit, I think the SCTP proposal is more
interesting.  Check draft-ietf-sigtran-sctp-06.txt.

							K. Poon.
							kcpoon@eng.sun.com