[OLUG] Collisions?

Vincent vraffensberger at home.com
Tue Feb 1 05:07:52 UTC 2000

> As for the reason you only saw 6mbit is because of the overhead of the
> protocals.
> 1) Ethernet packets themselves have overhead, checksums, hardware
>    adresses, etc
> 2) TCP has overhead, IPs, it's own checksums, etc
> 3) FTP has overhead, not sure what, but I'm sure there's something
> 4) packets get droped, no nic/hub/etc is 100% things get lost, that's why
>    there's all thos checksums.
> There's other stuff, but that's just an extremly high overview of why you
> only see 6mbit. Other things come into play too, like packet size. If you
> trasmit 3k packets, you have less header overhead, but you have to
> retransmit all 3k for a failed packet. with 1k packets you'd have more
> overhead, but only 1k to retransmit if something went wrong. Stuff does go
> wrong, nothing any of us can do about it.
> Anyway, get a 100bt hub, you'll like it.

TCP has extremely an high overhead.  Most systems are usually too
under-powered to really take full advantage of a 100basetx NIC.  There
is an interesting solution though.  I've been checking up on Silicon TCP
for nearly a year now and have no idea why it hasn't started to catch on
yet...  Here's a quote from their site:

...applications like file/web serving, file transfer, audio/video
transmission, and other data intensive operations on the Internet. Many
of these applications see as much as 90 % of the processor consumed by
Internet communications - if that resource was given back to run the
application, almost 9x the number of applications could be run with the
same processor. Since Silicon TCP scales with the bit rate of the link,
a gigabit product would expect to increase the performance gap still
larger. Even with expensive specialized NICs, fatter (8192 byte)
packets, and elaborate software modifications (NT5), one can only get
65-75MB/ sec with 30% of the processor. While with only a upgraded PHY
and wider datapath, the existing board design demonstrated can reach
90MB/sec with less than 5% of the processor. An even more exciting
design could allow 850MB/s on a 10Gbit PHY, where even megabyte sized
fat packets wouldn't be enough.

Do not meddle in the affairs of SysAdmins,
For they are subtle and quick to anger.

Sent by OLUG Mailing list Manager, run by ezmlm.  http://olug.bstc.net/ 
To unsubscribe: `echo unsubsribe | mail olug-unsubscribe at bstc.net` 

More information about the OLUG mailing list