On Wed, Dec 05, 2001 at 11:34:27AM -0800, kml@selresearch.net wrote:
> 
> If I can be so bold, the difference is that network nodes
> are supposed to work and play together nicely, according to
> well defined, published standards.  Hardware has always been 
> completely random, and, to me at least, scary -- what standard
> does any of it adhere to, other than, say, NE2000 cards?  I've
> yet to buy a network card (other than NE2000s) that promised
> me much at all about its register structure, and I've certainly
> been burned more than once by this.  TCP/IP means something, and
> folks oughta adhere to that definition.

Isn't SCSI supposed to be a well defined, published standard?  Yet
we have to work around broken implementations.

> I think that working around other people's frailties is a good thing,
> in general, but the MSS clamping stuff is pretty close to where I draw
> the line.

My whole point was that we work around other broken standards or devices,
and adding a feature to allow the end user to enable a work around for
the countless "broken" sites out there seems like a good idea.  Provided
the changes are clean and simple anyway.

I think it sounds like an especially good idea if we plan to tout our
in-kernel PPPoE implementation as a better alternative to RP PPPoE from
pkgsrc.

-Andrew