[c-nsp] Horrible MPPP Performance

[Michael K. Smith]

Hello Seth:


On 3/6/09 8:12 PM, "Seth Mattinen" <sethm at rollernet.us> wrote:

> Michael K. Smith wrote:
>> Hello Everyone:
>> 
>> I have two 2800 series routers with 4, clear-channel T-1's between.  I'm
>> running MPPP with the 4 T1's in the bundle.  Performance is *awful*.  100
>> byte packets, 40 ms with 98% delivered.  1500 byte packets, 900 ms latency
>> with 25% packet loss.
>> 
>> Here are my config snippets from the two routers.  I've done everything I
>> can think of and nothing seems to change the performance.  Any help would be
>> greatly appreciated.
>> 
>> I've tried moving clock from one side to another, tried running with
>> network-clock-participate on the line-clocked side, set both sides to line,
>> set both sides to internal.  The configuration below is the only one that
>> shows no slips, but the performance is still terrible.
>> 
>> I've also tried changing the load-balancing from per-packet to
>> per-destination to FIFO to Fair queue.  Nothing changes.
>> 
> 
> 
> Why do you have load-sharing with MLPPP?
> 

As far as I can tell, with CEF enabled, it's either per-packet or
per-destination, with per-destination being the default (not printed to the
config).  I tried doing the "no" on the per-packet with no change.

Regards,

Mike