MRTG averages over a time period - mean vs sampling

MRTG averages over a time period - mean vs sampling

Post by Michael L. Samue » Tue, 13 May 2003 21:41:48



I am new to MRTG, and I am trying to compare its results to some
packet traces recorded by Netmetrix.   The MRTG graphs sent to me show
"average" and "peak" utilization over a 1-hour period.   There are 14
data points - it looks like 1 at the beginning of the hour, 1 at the
end, and 12 in between, so presumably these stats are recorded every 5
minutes.

My question is whether the stats are based on raw data collected
during a 5-minute period or whether MRTG is just recording 1 packet
every 5 minutes.  I am assuming that it collects every packet over the
interval, determines the mean and max, and then starts over with a new
5-minute interval.  Is this correct?  (Netmetrix, when used in
"sampling" mode, actually only records every Nth packet, where N is
configurable and dependent upon the link speed.  That's why we use it
in packet mode, where it just records every packet it sees.)

Thanks.

 
 
 

MRTG averages over a time period - mean vs sampling

Post by Mats Karlsso » Tue, 13 May 2003 22:30:39




Quote:> I am new to MRTG, and I am trying to compare its results to some
> packet traces recorded by Netmetrix.   The MRTG graphs sent to me show
> "average" and "peak" utilization over a 1-hour period.   There are 14
> data points - it looks like 1 at the beginning of the hour, 1 at the
> end, and 12 in between, so presumably these stats are recorded every 5
> minutes.

> My question is whether the stats are based on raw data collected
> during a 5-minute period or whether MRTG is just recording 1 packet
> every 5 minutes.  I am assuming that it collects every packet over the
> interval, determines the mean and max, and then starts over with a new
> 5-minute interval.  Is this correct?  (Netmetrix, when used in
> "sampling" mode, actually only records every Nth packet, where N is
> configurable and dependent upon the link speed.  That's why we use it
> in packet mode, where it just records every packet it sees.)

> Thanks.

MRTG uses by default the snmp counters in the device you are pulling for
snmp data.

Or should I explain it like this: The snmp counters are a acumulated value.
MRTG is calculating the traffic as the difference betwen two sampled values
during a time period.

The snmp counters can be a 32bit or 64 bit value (depending on manufacturer
and snmp version).

Kind regards
Mats

 
 
 

1. MRTG problem: Blue lines average 5 samples, green is per sample

Everything works on mrtg great, but we were doing some throughput
testing and following the path of the test through 8 interfaces, all
monitored by MRTG on several switches

What I found, which I had suspected before was that the values that
plot the blue-Outbound traffic are not calculated correctly. They
appear to be averaged over 5-6 samples.

In our test I sustained a 50 megabit load end-to-end.

On the IN-green chart of each port involved the values were a straight
line, showing a sustain 50Mb transfer for one hour.

for example:
00:00  50mb
05:00  50mb
15:00  50mb
20:00  50mb
25:00  50mb
30:00  50mb
35:00  50mb
40:00  50mb
45:00  50mb
50:00  50mb
55:00  50mb

THE EXACT SAME TRANSFER on blue-outbound ports showed

for example:
00:00  8.mb
05:00  17mb
15:00  26mb
20:00  38mb
25:00  49mb
30:00  50mb
35:00  50mb
40:00  50mb
45:00  50mb
50:00  50mb
55:00  50mb

This "ramp up" is impossible, because the next hop showed in green a
50 Mb transfer during "ramp up".  Clearly there is a calculation error
occuring in the outbound levels. It takes at least 5 samples for the
outbound-blue graph to show actual levels.

Does anyone know of a bug or fix?
I'm checking our MRTG now. It was setup by a consultant--I actually
have no clue of the rev were running (on Solaris).

2. EPIA'95 - Conference CFP

3. MRTG quits after a period of time

4. GDI System Call Hooking

5. Average Handle Time vs Not Ready

6. FS: CNet Pro

7. MRTG and linux load average.

8. Driver for Packard Bell keyboard FDA-116 C

9. Using MRTG to monitor server load average

10. MRTG finding a true average util based on operating hours

11. MRTG: Changing Graph Averages

12. MRTG Average Traffic Statistics

13. Computing monthly averages using mrtg logs?