Re: profiles - a way to simplify iSCSI

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

> I think this is indicative of the complexity we have in
> the standard. The first order of preference should be
> to reduce the complexity.

I agree with Somesh.

When we talked to (some of) the FCP gurus when starting SST, their
single most prominent piece of advice was to avoid options and avoid
profiles.  Seems like motherhood to me.  Profiles usually end up being
ways of pruning features that a vocal minority thought were critical,
and a silent majority didn't really care about and didn't plan to
implement.  It really makes it hard to figure out what you have to
implement, and that's really what happened with FCP.  Then again,
maybe such surprises are in somebody's interest (heck if I can figure
out whom).

Personally, I think we're much better of starting out with a
rock-solid, interoperable first step, and extending from there, than
starting with a specification that covers many anticipated
possibilities that we may get around to eventually.  The IP protocol
suite was developed along those lines (start with something basic, and
then tweak and enhance), and it's worked out well.  SCSI itself
followed that model (ah those grand days before queuing, and single
drop links).  OSI followed the architect everything before we
understand approach and, well at least 2 of those layers fell unused,
not to mention the protocols themselves.

Nonetheless, if y'all really think the spec has to option laden, it
has to be option laden, and NO profiles.  If we make this mess, we
should lie in it.

Steph