RE: iSCSI Markers

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

Paul,

I agree with you and John.  Maintaining conformity as well network layering
is important and advantages offered by the new transport easily outweigh
concerns balanced against modification of TCP.  No one needs to ask my
preference. : )

Doug

> >>>>> "John" == John Hufferd <hufferd@us.ibm.com> writes:
>  John> You have FIM or COWS, neither one is useful with TUFs, except
>  John> that they have a similar header with COWS, not is hardly a
>  John> reason to put up with the SCANNING and Stuffing of COWS
>  John> Pointers in the Data.  This is especially since depending on
>  John> the development approach you may find the Pointers pointing the
>  John> wrong way for either sending or receiving in some HW pipelining
>  John> designs.  And it is always a poorer approach then FIM in
>  John> Software.
>
> Agreed. FIM is nasty, COWS is nastier.  Both may require a memory to
> memory copy, but at least FIM doesn't require you to examine the data
> you're moving.
>
> Clearly, the real answer is a transport layer that isn't a byte stream
> transport.  TP4 or SCTP, anyone?  :-)
>
>        paul
>
>