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

["Justin T. Gibbs" <gibbs@freebsd.org>]

>   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.
>
>With client memory bus bandwidth in the multi-gigabyte per second
>range, who needs to avoid the single copy?  How much NFS and web
>surfing does one need to do before this is would really come into
>play?

This is a very different argument than the one you "implied" before.
I can certainly say that I'd rather make use of my memory bandwidth,
regardless of how much I happen to have, in a more constructive manner
than copying data for no good reason.

RDMA is a general purpose feature.  Don't shoe horn it into just an
option to accelerate a few protocols.

>And the bus speeds will just be faster by the time something like
>this could be deployed widely.

And the networks will be faster by then too and the demand for
pulling more rich content will increase, etc.

>For example, look at SACK, only within the past year are there a
>decent number of systems out there implementing it.  Now how many
>years ago did it enter RFC state?  And there are still stacks out
>there even in their current development sources not implementing any
>form of it.

The performance impact of RDMA is quite a bit larger than SACK, so
I don't know that your example is relevant.  All the big vendors
implement zero-copy in some shape or form and since RDMA is a scheme
to make zero-copy work in more cases, I'm sure it will be picked up
if the proposal is deemed sane.

>Later,
>David S. Miller
>davem@redhat.com

--
Justin