On Tue, 24 Oct 2006 10:38:49 -0700, Cliff Wright <cliff@snipe444.org>
wrote:

> With all this talk of hardware checksum lately, I thought I would
> mention their was a recent discusion (which I could not find) of
> the results of a University study into packet errors and why they
> occur. One conclusion of this study was that hardware checksum
> was not that useful as it just backed up the CRC, however that
> software checksuming was very useful and would many times catch
> the somewhat rare error that is becoming more common as very large
> file transfers become common. They also concluded that not all errors
> can be found, and that for critical data the app will have to do a check.
> I leave the software check on so that I don't have to always think
> about what kind of network activity I am doing.
> 
You're probably thinking of "When The CRC and TCP Checksum Disagree", by
Jonathan Stone and Craig Partridge
(http://citeseer.ist.psu.edu/stone00when.html).  It's a classic paper.
(And note that Jonathan is a NetBSD developer.) 


		--Steven M. Bellovin, http://www.cs.columbia.edu/~smb