RE: Urgent as Framing Hint?

[Douglas Otis <dotis@sanlight.net>]

Matt,

Should you consider encapsulating FC frames, using a maximal frame as a
fixed interval, this would place the header interval at roughly a 2k
boundary.  You could reject use of optional FC header extensions not used in
FC Class 3 SAN.  FC frames would also allow a simple method to regulate
buffer-to-buffer cache allocations.  No mode pages or cache methods would
need to be changed as advocated with iSCSI.  The document I prepared should
illustrate direct use of the FC Frame is not significantly different than
the iSCSI PDU.  With this, there is no frame discovery problem as it becomes
a simple matter of division where modulo conditions are considered should
this interval not be a power of 2.  Should connection nexus or other
information be found necessary, then add this information as a prefix as
illustrated.  All high speed processing of network information will coalesce
interrupts so smaller frames do not represent a burden as this is handled
within hardware.  The header overhead is also compensated by higher
bandwidth made available by 10 Gbit networks.  What is the overriding desire
for reinventing the typical SCSI serial frame?

Doug

> Rick Jones wrote:
>
> > if _every_ iSCSI PDU were always 1024 bytes (some fixed size) you would
> > know where they were in the TCP datastream all the time.
>
> This would be very wasteful of bandwidth for commands and status.
>
> -Matt
>