[CentOS] SNMP and OID/MIB/MRTG

Fri Sep 12 03:40:30 UTC 2008
nate <centos at linuxpowered.net>

Bob Hoffman wrote:
> You know...it would be nice to write a program that would use snmp to just
> scan through the local computer, grab OIDs...and then ask you which ones you
> wanted to make MIBs with.
>
> Then you could just grab those with the graphing programs.
> I think that would be an awesome tool to build for linux.
>

It's not exactly free and it's not web based, but this is a pretty
good tool, I plan to buy it pretty soon(been using the free
version off and on for a while now. though it's limited).

http://www.ireasoning.com/mibbrowser.shtml

I've been digging quite a bit deeper into SNMP the past couple
months at my new company monitoring hundreds of stats from our
network equipment, and having the mibbrowser is really really
helpful. No way I could of done most of it without it. My
cacti system collects more than 10 million data points a day,
on one dual proc quad core box. More than 95% of the stuff I
put into cacti comes from scripts I wrote(I write the
scripts to gather many data points simultaneously to reduce
the amount of RRD data files stored improving performance by
more than 10x.)

If someone knows of a better MIB browser I'd certainly be open
to checking it out, having looked around quite a bit the past
couple years I haven't found anything better myself that runs
on Linux.

I certainly do agree that SNMP is a black art, not sure why
it is so complicated, perhaps it just helps to sell those
$100k enterprise monitoring packages because there's little
hope for the average admin to figure out how to do it on
their own.

At my current job(started in March), before I came on they
were telling me how the previous admin setup CPU monitoring
in cacti, and yet the CPU graphs never seemed to go above
25%. They weren't aware that the CPU usage reported by the
snmp daemon used in linux returns useless, completely
inaccurate data(this is documented pretty clearly in the
daemon documentation but doesn't seem to be common
knowledge).

I've refined my data collection scripts over the past 5
years or so, they work great now. CPU usage for my cacti
systems is sourced from 'sar'. I really hate how sar has
gone down hill as far as ability to parse it. RHEL 3
was great, RHEL 4 was ok, and RHEL 5 is almost useless,
don't know what I'll do when RHEL 6 comes out.

nate