[c-nsp] MLPPP maximum load

[Adrian Chadd]

On Mon, Aug 13, 2007, Robert Boyle wrote:
> At 10:38 PM 8/13/2007, a. rahman isnaini r. sutan wrote:
> >Am I right to say that MLPPP could utilize all bundled T1/E1 100% ?
> >Or it might be wouldn't work normally as there some headers added on ?
> >And recommended is 75% for a normal load balancing.
> >
> >Any suggestion would be appreciated.
> 
> The MLPPP line protocol overhead is negligible. However, on some 
> platforms, the CPU overhead is high. On any modern Cisco router 
> (18xx/28xx/38xx+) or any higher end older router (3700,7200) you 
> should be fine up to the maximum of 8xT1/E1. Try 8xT1 with a full 
> 12Mbit/s on a 3620 or 4xT1 on a 2600 and you may run out of CPU.

It becomes less negligible as you add more bundles - it splits the
packets up and transmits the fragments in parallel. Smaller packets ==
larger overhead == more waste. There's a diminishing return on
adding more bundles into MLPPP.




Adrian