Re: [nsp] bit rate accuracy?

From: Gert Doering (gert@greenie.muc.de)
Date: Tue Nov 13 2001 - 14:40:48 EST


Hi,

On Tue, Nov 13, 2001 at 12:30:51PM -0500, Deepak Jain wrote:
> If I remember correctly, Cisco uses some kind of decaying average algorithm
> to report bit rate/second on interfaces.

Yep.

> MRTG and programs like it use regular polling of byte counters. If the
> latency of the network is constant and both the statistics machine and the
> network device have sufficient available CPU, it seems to me that MRTG
> statistics will be more accurate over a 5 minute window.
>
> The reason I am asking is because I have a full duplex FE interface that is
> reporting 94mb/s inbound on a 5 minute average, and yet the MRTG is never
> exceeding 84.2mb/s on the same interface.

It always depends which bugs you're hitting and over which time you're
averaging...

We're reading the SNMP octet counters, the "show int" octet counters, and
both SNMP and "show int" 5 minute average values. All of those values
are read every 5 minutes, averaged over the day, and then compared, and
the results differ wildly (interesting enough, SNMP and "show int" values
differ by up to 20% for the 5 minute average counters - interesting,
isn't it?).

The most exact seems to be the SNMP octet counters so far, especially on
high speed interfaces when you have counter64 available (and if they
happen to work).

The problem with the octet counters is that they are not "so nice to look
at" - with the 5 minute averages, you can just look at an interface and
see what's going on. With octet counters, you always have to calculate
the deltas - easy for a machine, cumbersome by hand.

So both types of counters have their merits.

> [MRTG is configured as a daemon so cron-timing shouldn't be an issue]

Actually if you sample every 5 minutes, and note down the exact time when
the sample was read, it doesn't really matter whether one sample is
delayed by a few seconds (or even minutes) - you have to know the time
difference between the two samples anyway to calculate the average, and
if you query the router later, the counter will be accordingly larger...

gert

-- 
USENET is *not* the non-clickable part of WWW!
                                                           //www.muc.de/~gert/
Gert Doering - Munich, Germany                             gert@greenie.muc.de
fax: +49-89-35655025                        gert.doering@physik.tu-muenchen.de



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:23 EDT