RE: iSCSI: Towards Urgent Pointer Consensus

["Douglas Otis" <dotis@sanlight.net>]

David,

It is not unreasonable to request a fuller picture of intent. I think you
will find my concerns similar to Randall's with respect to the single Urgent
Pointer used in the stack according to the TCP specification without looking
at implementations.  With that said, I suspect those advocating use of the
Urgent Pointer are not concerned about TCP limitations.  To move past the
areas of concern Randall has observed, next would be how this mechanism will
be used assuming its generation is not an obstacle.  I can become creative
about how it may be used, but those proposing such a mechanism owe the group
their concept.  If I decry a problem, my imagination may not be as clever as
yours or theirs and thus waste time and incur the ire of the chair.  As a
general guideline, implementations which impact normal application API
should provide an outline of the revised API.

Doug

> Douglas Otis wrote:
> > Without at least an outline as how an API is to work which
> benefits from the
> > use of the Urgent Pointer Record Marking scheme, it is
> difficult to properly
> > discuss relative merits.  We can all imagine such an interface can be
> > created, but until details are roughly specified for the API to at least
> > proximal levels found in TCP or SCSI, then again, I think it is
> premature to
> > spend effort discussing this imagined mechanism.
>
> Again, I think URG will fail for other technical reasons, but defining
> an API to use it is not rocket science. Given that I am not very
> supportive
> of the use of URG I am not going to was the 15 minutes it will take to
> define an API.
>
> If you want to defeat URG, use technical arguments like Randall has been
> doing.  Please don't go down misguided argument path about APIs.
>
> 	-David
>