[nsp] atm encapsulation overhead

[Iva Cabric]

On Tue, Nov 18, 2003 at 10:26:45AM +0100, Gert Doering wrote:
> Unfortunately, that's not all of it.  With aal5mux, a TCP ACK packet with
> no data fits into *one* ATM cell, with aal5snap, you will always need two.
> 
> So for typical web traffic (data going one way, pure ACKs in the other
> direction), aal5snap overhead is significantly higher.

Certainly, and it applies for any other type of traffic, VOIP is also
good example, because choosing wrong payload size can lead to spending
one more ATM cell per packet, and on lower bandwidth links, one needs to
use LFI, which also has its own overhead (for MPPP over ATM), and so
on... (more on that in
<URL:http://www.cisco.com/warp/public/121/lfi_24041.html>
)

> I don't have numbers, though.

It is not even possible to have real numbers, it all depends on type of
traffic which goes through link.
And this should be an answer to Dan's question in first mail, there are
no simple numbers, you need to analyse your traffic and then apply
overheads from lower protocols.

But keep in mind that if average size of packets is, for example, 500
bytes, your real traffic can have all other sizes but this one, and
your expected throughput will be higher then real one.
In worst case you will lose one ATM cell per packet when payload (IP)
is one byte too large to fit in expected number of cells.