Re: iSCSI: draft 7: remove S bit and Status field from the Data-in PD U?

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

Robert,

> A similar simplification would be to remove the S bit and Status
> field from the Data-in PDU.

It wouldn't hurt my feelings if this happened.

However, there is a difference between these two simplifications.  The
good status optimization prevents the transmission of a PDU which is
~100% overhead.  Unsolicited data, by its nature, is to carry data, so
adding additional headers just leads you to the current header/data
overhead level.

That said, the ratio of data to status is going to be high no matter
what, so the overhead contribution of not optimizing good status is
probably neglible.  Furthermore, for hardware implementations, the
good status optimization is just an irritant.  Presumably that is why
FCP never bothered with it.

The traditional-sockets software-only implementations will have the
same argument against eliminating this optimization that they do
against removing Immediate data.  These optimizations permit them to
do half as many reads (2 instead of 4) in the optimized cases.
Personally, I would prefer not to give weight to unoptimized TCP
software solutions over hardware and optimized TCP software solutions.
Although most existing implementations are unoptimized TCP versions,
they should be considered prototypes (after all, version=0 until RFC
or death :^)

At an absolute minimum, I think we could remove the status field, and
assume that s=1 implies status=good.  There are no other SCSI statuses
that are worth optimizing.

Steph