[c-nsp] SNMP: cacti graphing of 6500 interfaces more accurately

[Jon Lewis]

On Wed, 1 Feb 2006, Tristan Gulyas wrote:

> Currently I am using a built-in cacti template for interface statistics
> which I believe bases itself on a 5 minute average counter.  We have another
> network monitoring system (Statseeker) which seems to give us substantially
> clearer results (not only does it poll every minute bit it seems to show
> traffic spikes more accurately).

You're probably running cacti's poller every 5 minutes.  SNMP presents raw 
counters...not 5 minute averages.  If you poll the counter every 5 
minutes, all you know is the counter went from X to Y in 5 minutes.  Was 
it a constant Y-X/300 bit/s, or was there a relatively large spike 
somewhere in that 5 minute window?  You have no way of knowing.

If you want better resolution in your graphs, you have to poll more 
frequently.

BTW, the [old?] default poller in cacti (not the cactid poller) is kind of 
slow, and if you have a lot of hosts/interfaces in cacti, you're probably 
going to run into problems, especially if you try polling more frequently.

----------------------------------------------------------------------
  Jon Lewis                   |  I route
  Senior Network Engineer     |  therefore you are
  Atlantic Net                |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________