Why no stats.tx_bytes / stats.rx_bytes

[Donald Becker]

On Thu, 5 Nov 1998, Ricky Beam - Interpath wrote:

> > Well, mostly correct: we haven't defined what the tx_bytes/rx_bytes fields
> > should count.
> 
> IMO, {t,r}x_bytes should count the number of bytes that are flying over the
> cable - period.

Errmm, including the hidden 8 byte preamble?  Probably not.
The "correct" count depends on your layering model.  Most people use the OSI
layering model to describe networks, but OSI doesn't map well to reality?

[[ OSI was a mostly-European effort to redefine networking.  Like ATM, it
was promoted by organizations that had fallen behind or made bad choices and
wanted everyone to start from scratch.  It was composed of complicated,
committee-design protocols that had never been implemented and were horribly
broken.  Very little of OSI remains today except for the layering model. ]]

> > We always omit the 4 CRC bytes that are added and stripped by the hardware. 
> 
> EEPro100's do hardware CRC?  Shit, I'm liking these things more and more
> everyday!  (beats the crap out of the 3c905's; of course, a roach squished
> into an ISA slot makes a better NIC that a 905.)

Every card does hardware CRC.  It's UDP/TCP/IP checksumming that's rare.
The 3c905B "Cyclone" chip core (but not the previous generations) actually
has the best implementation of IP checksumming of any design I've seen.
The current EEPro100 has no IP checksumming support at all.

One problem with the 3c905B appears to be that 3Com uses the MII transceiver
design provided by the fab line the chip is being produced on.  This way
3Com can purchase chip production time from whichever fab line is least
expensive at the time.  But not all of the MII transceiver designs act the
same.. 

Donald Becker					  becker@cesdis.gsfc.nasa.gov
USRA-CESDIS, Center of Excellence in Space Data and Information Sciences.
Code 930.5, Goddard Space Flight Center,  Greenbelt, MD.  20771
301-286-0882	     http://cesdis.gsfc.nasa.gov/people/becker/whoiam.html