[In the message entitled "Re: NFS lossage from NetBSD to SunOS" on May  2, 23:04, Valtteri Vuorikoski writes:]
> On Thu, 2 May 1996, Gordon W. Ross wrote:
> 
> > Is there a good test to determine when this lossage exists?
> > BTW, Why is the Ethernet CRC not a sufficient error check?
> 
>  At the moment netstat -s on the SunOS box is showing
> 
> udp:
> [...]
>         7 bad checksums
> 


There are many reasons this can happen - which is one of the reasons
that I strongly suggested second level CRC's be used on the PPP CCP.

The ethernet hardware generally includes some form of buffering.  
The ethernet CRC-32 happens before this buffering.  Several 
race conditions can exist, causing flawed access to this data.
On the National Semiconductor early parts, an even more subtle
error caused one or two bytes, on random packets, to be corrupted.

It costs very little to check data when crossing CPU boundaries, and
it saves many, many headaches.


-- 
Dave Rand
Internet: dlr@daver.bungi.com