[mrtg] Re: bandwidth utilisation

[Rich Adamson]

> > That may not be totally correct. The RRDtool can poll remote devices
> > at a much faster rate. If the tool is polling faster than what the
> > counters are rolling over, it could be beneficial. 
> 
> While your statement about RRDtools ability to poll more frequent
> is technically correct, so is my statement about 2^32.
> 
> Buying faster boxes with more memory, cpu, bus and so on than you and
> I can afford will also allow MRTG to poll devices more frequently.
> 
> A big problem with polling more often is that, at least this is my
> opinion, the error introduced in the measurement increases.  If you
> have a timing error of 1 second every 300 seconds vs. the same 1-second
> error every 60 seconds then

I agree with you and the snmp community effort to define 64-bit counters,
etc.

I've been doing network performance reviews for enterprise clients in
40+ US states over the last eight years. We've basically eliminated the
polling error issue that you mentioned by including an snmp oid request
for the device sysuptime in "every" snmp poll. In order to ensure some
reasonable level of accuracy (in this very statistical net world), our
internally developed software requests the sysuptime, octets in and out,
pkts in and out, discards, errors, etc, all in _one_ snmp get. The timer
is accurate to within a millisecond or two and removes transient time,
snmp management system overhead, and many other errors from the 
inaccuracies that you mentioned.

In corporate networks that handle a lot of response-time-sensitive data
flows, we also poll remote devices on either a 10 or 15 second basis in
order to "see" peak periods that impact interactive sessions. In the
mostly non-interactive networks, we'll increase the polling rates to
60-seconds or so. Our timing errors are on the order of 0.1%

Since there are not a lot of servers in the corporate environment that
can _actually_ run any respectable utilization on a gigabit interface,
32-bit counter roll-overs in a ten second poll are infrequent at best.
I'm sure that will change over time.

Using a five-minute polling rate is fine for trending and lots of other
purposes, but is far too large for identifying significant anomalies
impacting the business environment. 


--
Unsubscribe mailto:mrtg-request at list.ee.ethz.ch?subject=unsubscribe
Archive     http://www.ee.ethz.ch/~slist/mrtg
FAQ         http://faq.mrtg.org    Homepage     http://www.mrtg.org
WebAdmin    http://www.ee.ethz.ch/~slist/lsg2.cgi