Interface stats!

[dlaperle]

Hi boys!

I'm using Cacti since 2-3 months...i really appreciate this software and your help, but now i'm in front of a problem that can only be solve by you...

Yesterday, we did a test on a link to see if Cacti take the right data.

switchX -- switchY

We had send a 1Go file between two servers behind these two switch!

It take 30sec and our Bandwidth Gauge (Solarwinds Engineer Edition 2k5) indicate 175mbps...

The switch port indicate a 5min average of 64mbps...

Is this possible for Cacti to base his formula with the EXACT data capture on the port when he polls?

I don't know if i'm enough clear, but i hope so cause it's a big problem here in our network...cause all our graphs arent really true...we have a port that go at 200mbps and Cacti show me only 115mbps cause of the average...

Thks for your wonderful support!

[coffee]

I just ran into a similar problem. If its the same, I'd say your using snmp v1 instead of 2c and 32 bit counters. Change the host profile to snmpv2c with 64bit counters and it should work

[ymilner]

Cacti (and any other similar tool) calculates load, based on number of bytes passed through interface. To do so, it gets this number once, then in the next poll time it gets it again, and based on those two readings it calculates the AVERAGE speed during this interval (5 minutes by default). If your file transfer lasts long enough, cacti will show the load that is very close to the real-time load on the interface; otherwise you have to live with average.

The problem here is, that the switch (or other device, for that matter), usually does not store information about the real-time load, at least in a way ir can be read by Cacti. When you see the load statistics on the switch itself (cisco, I assume), it uses exactly the same method - having 5 minutes as it's interval. At the switch itself, you can change this interval to be as small as 30 seconds, and then what you see at the switch will be more realistic. But to do the same in Cacti, you would have to change the polling interval to the same 30 seconds - which is, well, "not so good".

[dlaperle]

Thks guys for your reply!...

All my network is in snmp v2c and soon v3...

I think i'll let the graph like they are, cause i don't want to to submerge my network with flow of polling because we have a lot of monitoring console that poll ...

- CiscoWorks
- Cacti
- Nagios
- Unicenter CA

So, if i change the polling options of one system, i absolutely need to review all the polling in the others systems cause if i don't do so, i will have too many polls at the same time on my network..

I'll keep cacti like that and tell everybody that is an average and not the REAL data...so i hope they will understand!

Thanks for your time!...