[Gllug] OT: MTU sizes, w/ Broadband

[Simon Wilcox]

On Wed, 14 May 2003, Chris Bell wrote:

>    The problem is that BT have been using their ATM network to carry the
> data. Their network was specifically designed for speech using two parallel
> data streams, one carrying route data and the other digitally coded
> real-time audio, using very small packet sizes. It was probably very
> advanced at the time it was designed, perhaps it was a work around when many
> large but slow and simple chips were required, although few people
> understand the reasons now. The optimum MTU for this system works out at
> 1458, until they can install new equipment designed to work with an MTU of
> 1500, although the affect is minimal unless their system approaches
> saturation.

ATM uses fixed length packets, called cells, which allows highly efficient
aggregation of data on backbone networks using well understood, and very
fast, time division multiplexing techniques. It also includes quality of
service metrics and other features useful for audio/video and other time
senstive packet systems.

ATM cells are 53 bytes long, with 48 bytes of user data. The ethernet
packet is chunked up and reassembled at the other end. Under length cells
are padded to the exact length. So you can gain *margin* increases in
speed by making the MTU of your ethernet such that you exactly fill the
last cell thus saving the padding.

Personally, I've always considered this to be so marginal, given the 
general latency of the internet, to be not worth bothering with. YMMV :)

Simon.

PS. ATM use is generally considered a feature rather than a problem as it 
enables very efficient, and therefore cheaper, backbone networks. Ethernet 
isn't very efficient but as the speeds have increased, it's cheap 
componentry and the glut of fibre in the ground has meant that it is used 
over longer and longer distances these days.



-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug