Re: Some Thoughts on Digests

[Matt Wakeley <matt_wakeley@agilent.com>]

julian_satran@il.ibm.com wrote:
> 
> 
> Hardware implementation complexity is not a good counterargument as long as
> it does not translate into a larger silicon area -:)

The way it is currently defined, it requires a lot of silicon area!  You want
a separate CRC-32 per 2K block.  Since the CRCs are accumulated during the
transfer of the payload, and then tacked on the end, there must be storage in
the silicon to store these values.  Likewise, when receiving, the CRCs are
accumulated and then must be verified at the end. So, if sending a 256K iSCSI
PDU, you must accumulate 128 CRC-32s and store them somewhere (most likely in
registers), and then tack them on the end of the PDU.  The larger the PDU, the
more CRCs must be stored.

-Matt

> 
> Regards,
> Julo