[j-nsp] EX3200 Odd behavior on cacti

[Marco Matarazzo]

Hi Bill,

> I am currently using a EX3200 in a routing capacity. At any given time it receives some good PPS bursts. I've graphed it at 100k PPS and 150-200M on my upstream interface. Should be nothing for
> a line rate device. However I'm seeing a drop in graphed bandwidth that consistently coincides with a high PPS count on my cacti graphs. I try to capture an event using 'monitor' but haven't yet. My
> interface counters show no errors or problems. What I may be experiencing is a problem with SNMP or with cacti even though my cacti instance doesn't follow this pattern for other devices. I would
> ike to verify if traffic actually drops or is SNMP gets stupid when a high number of packets are blasting through the router.

Make sure you're graphing 64bit counters in Cacti. If you're not, the
graphs will 'rollover' beyond 120Mbps per minute, thus giving the
impression that the bandwidth is actually going down instead of up!

> Since cacti uses 5 minute trends, in order for something to even show up (extreme bandwidth drop) it would have to be very low for most of that period.
> Is there a way to log interface stats in near real time? At least the high and lows in an attempt to correlate these events.

You should be able to go down up to a minute with recent versions of
Cacti, and you can do subminute samplings with rtg
(rtg.sourceforce.net).

Cheers,
]\/[arco
-- 
I'm Winston Wolf, I solve problems.