[nsp] PPS Interface Counters

[Ryan O'Connell]

Lawrence Wong wrote:

>Thanks for explaining how the pps is calculated. I
>assume this is averaging out all the pps on all
>interfaces to have a rough guage of the average pps
>throughput?
>  
>

It's because each (normal, unicast) packet is accounted for in the "show 
interface" statistics twice - once on it's way into the router and once 
on it's way out. If you then sum up all the pps counters, every packet 
is counted twice so you need to divide by two to get the right figure. 
Obviously this breaks if a packet comes in but doesn't go out (Broadcast 
not forwarded by the router or multi/unicast destined for the router 
such as ping, OSPF etc) or is forwarded out multiple times (Multicast) 
so it's not 100% accurate, but should be pretty close. (Certainly within 
5%, although bear in mind this is just a snapshot and peak pps doesn't 
necessarily coincide with peak bandwidth so graphing over time is good 
if you're using these figures for capacity planning)

>Speaking of MRTG, do you know of any OID for Cisco
>which shows say maybe the 5 minute average pps ?
>
>I tried to view .1.3.6.1.2.1.4.6.0 and it shows an
>ever increasing counter.
>
>IP-MIB::ipForwDatagrams.0 = Counter32: 3650827250
>IP-MIB::ipForwDatagrams.0 = Counter32: 3650828792
>  
>

As far as I know, the only way to get the 5 minute averages without 
polling the above every 300 seconds is via "show interface". If you put 
the above in MRTG/Cricket as a counter - the same as interfaces - you'll 
get the right 5min average out however.

Personally, I use Cricket with genRtrConfig (Google will find you a 
copy) to get all this information both on a per-chassis and 
per-interface basis.

-- 
         Ryan O'Connell - CCIE #8174

I'm not losing my mind, no I'm not changing my lines,
I'm just learning new things with the passage of time