Ali Khayam <ali_khayam@yahoo.com> writes:
> > > > Of course it can. It is totally trivial. None of
> > > > this need be done in
> > > > the kernel, or at the UDP layer.
> > > 
> > > which means over-burdening the application layer.
> > 
> > Inserting two bytes is an overburden? Come on. You
> > don't even need the
> > header checksum in the first place, but if you want
> > it is burdensome
> > to insert two bytes and check them later? That's
> > ridiculous.
> 
> Its not the two bytes, its the processing overhead.

The processing overhead is a couple of clock cycles. The header is
static so it only has to be computed once.

> You also mentioned it and I also quote:
> 
> "the largest bottleneck to achieving high throughput
> in this system is due to computing checksums" 
>   J. Kay and J. Pasquale. Profiling and Reducing
> Processing Overheads in TCP/IP. IEEE/ACM Transactions
> of Networking, 4(6), December 1996. 

That's in TCP. You're mixing apples and oranges.

I think you really just plain aren't thinking about this problem
clearly. Luckily, the IETF has pretty strong mechanisms to prevent
such ideas from ever being put in the majority of implementations.

--
Perry E. Metzger		perry@wasabisystems.com
--
NetBSD Development, Support & CDs. http://www.wasabisystems.com/