RE: iFCP CRC

[Joshua Tseng <jtseng@NishanSystems.com>]

Wayland,

One important detail which was accidentally omitted from the
iFCP document is that the FC CRC (along with EoF and SoF) is
stripped off and not encapsulated in the iFCP PDU.  As you guessed,
the reason this was done is because iFCP changes the S_ID and
D_ID in the original FC frame.  The motivation behind adding it
back in at the iFCP layer is that some reviewers have expressed
the concern that the TCP checksum is not strong enough to catch
all bit errors. We will be adding the payload CRC as an optional
parameter declared during the N_PORT session setup.  That is, each
iFCP gateway indicates if the payload CRC is present by setting a
flag in the CBIND message.  If the destination gateway doesn't
support CRC, it can simply ignore it (skip the 32-bits) and rely
on the TCP checksum.

Thank you also for the comment that a header CRC might be overkill
to protect just the iFCP encapsulation header.  We will consider
making this a parameter declared during N_PORT session setup.
That is, each iFCP gateway will indicate in CBIND if a CRC, checksum,
or nothing is present to protect the header in the iFCP PDUs
that it originates.

Josh

> 
> Josh,
> 
> Charles mentioned in his presentation that a CRC was being 
> added to the
> encapsulation scheme.
> I'm wondering if you could elaborate on the motivations for this move.
> Certainly, It would be very
> desirable to protect the iFCP encapsulation header, but 
> adding another CRC
> could be costly
> for hardware. I know that "implementations" may be able to 
> ensure that FC
> traffic is entirely
> encapsulated within a single network MTU, but generally 
> speaking, that might
> not be the case.
> Requiring yet another CRC which is hard to manage across 
> network packets
> seems like overkill.
> 
> Fibre Channel and iSCSI both have considerations for CRC. I 
> know that iFCP
> mucks with the
> encapsulated payload (D_ID's and S_ID's), but it seems like 
> it would be
> completely acceptable
> to check and re-calculate the FC CRC to cover the modifications to the
> payload of the iFCP PDU
> (you have to do this anyway). To protect the iFCP header, I 
> would vote for a
> simple checksum which
> just covers the header. 
> 
> -Wayland
>