RE: iSCSI: reference for CRC32C?

[pat_thaler@agilent.com]

Luben,

Actually, I have referred people to that document. I think I have done it on IPS and I know I did it during discussions on SCTP. I have also referred some IEEE 802 people to it when they came to me with CRC questions.

However, I don't think the document is reference-able by an RFC. It isn't an RFC and it hasn't been formally published. To be an RFC it would have to be in text format. No one has been willing to take on conversion to text format which in any case would reduce readability as the equations get really long in flat text. When one cites a published article, then one can count on it being retrievable for the life of the RFC. It is in databases that can be accessed by libraries. Unpublished documents on websites don't tend to have that longevity. They are useful for discussions during development of a draft but they may not hang around for the life of the standard.

If you want formal references to it, I suggest that you and Vince get it published somewhere.

Regards,
Pat

-----Original Message-----
From: Luben Tuikov [mailto:luben@splentec.com]
Sent: Tuesday, July 30, 2002 12:57 PM
To: KRUEGER,MARJORIE (HP-Roseville,ex1)
Cc: Ips Reflector (E-mail); David Black; Julian Satran
Subject: Re: iSCSI: reference for CRC32C?


"KRUEGER,MARJORIE (HP-Roseville,ex1)" wrote:
> 
> Looking thru the current draft I can't find any doc. reference for the
> definition of CRC32C.  Does the text in sec 11.1 suffice, or is there some
> other document that fully defines the algorithm for CRC32C?  If so,
> shouldn't it be referenced in the bibliography?

If I read your sentence correctly ``other document that
fully defines the algorithm for CRC32C'', yes, there is:

http://www.haifa.il.ibm.com/satran/ips/Vince-Luben-crc32c-01.pdf

It fully ``defines the algorithm for CRC32C''.
It also quotes: (i) Castagnoli et al. which don't talk about 
algoritms but analysis of the CRC space, probabilites etc
and (ii) a paper by Williams which shows C code.

This is from the abstract:
	``The CRC32C (aka CRC32/4) digest from iSCSI is presented
in a rigorous algebraic manner, the why and how it works and the origin of
its verifier constant. The most commonly used CRC digest computation algorithm
in iSCSI and Ethernet, the Simultaneous Multiply and Divide (SMD), is derived
from the long division algorithm. Sample implementations are provided of
both algorithms.''

At the end, a pseudo code implementation is given of SMD, which is
what you'll find in actual implementations, including OS networking
layer code, network interface device driver source code and hardware.
The jump to SMD with table lookup (a great speed up) is trivial
and is also mentioned.

BTW, this exact question I've seen several times on IPS, ever since
that paper appeared, but no one has quoted it yet.

The irony is that, unbeknownst to me until last week, if you search
for iSCSI on google.com, on most of the cites that will come up,
you'll see several white paper references on iSCSI and one of them
will be ``the definitive guide to iSCSI CRC <<link to this
paper on Julian's web site or a local copy>>'', yet no one ever
mentions it on IPS...

-- 
Luben

P.S. Perhaps a Linux outlook on how science is conducted by
big companies is the only salvation. Yes, I know its new
and has never existed but in philosophy, only until recently,
but is more productive and fruitful. OTOH, everything I've
seen so far is exactly as History of Science would tell us.