Re: iSCSI: Framing Steps

[Stephen Bailey <steph@cs.uchicago.edu>]

> While I support a generic direct data placement model,

While you know that's the model I like, it isn't specifically what I
was talking about.

My only point was to enforce some separation of concerns between
iSCSI, which is basically an application protocol to carry the
RPC-type interactions defined by SAM, and how it is carried.

Agreed that TCP has `performance problems' at high speed (AI and MD
diverge).  Your following point:

> 5. A solutions that is iSCSI specific is not likely to give
> the savings (especially when it may not work) people expect.

hits it on the head.  TCP (or whatever transport) needs to solve this
problem in a general way.  The reason Ethernet/TCP/IP is economical is
because its approach to data transport is scalable and general.  The
various dimensions of cheapness start to evaporate when you say `we
just need this special hook to get application X to work well'.

I'm am so down with not specifying ANY framing in iSCSI.

Steph