[PLUG] Monitoring Unix systems with open source

[Eli Stair]

I just spent about 4 months doing some real hard testing, evaluating
(testing in production, writing tools to manage, integrating with
other apps) Zabbix/Zenoss/Hyperic/someothers to replace a
Nagios/Ganglia/Cacti system I've built over the last few years.  IMO
in summary, if you want to do anything more complex/custom than the
out-of-the-box features, Zabbix is the best bet (and is period).
Hyperic is closed-source in part and isn't very flexible, Zenoss is
great in theory, but changes require __LOTS__ of coding for the
middleware (Zope), and most of the features really are half-baked, and
most bugs were slow to get resolved (if ever).

Zabbix is the most extendable in a variety of ways; while I dislike
writing PHP, it is utterly simple, extensible, and fairly
flexible/capable (and only runs the frontend), support for SQL
backends is great and good for integration with external tools.
Zabbix flat out kicks ass as far as extensibility (able to arbitrarily
add SNMP/agent/script -based collection), custom collection methods
via any scripts on any platform (even Windows!).  Only problem is in
really large setups (thousands of monitored systems) which most people
probably won't run into... while I've spent over a month
isolating/fixing bugs with the Zabbix developers, its' still insanely
resource-intensive.  They HAVE been unbelievably responsive in fixing
each and every bug or performance/scalability issue I've brought up.

Even better, zabbix isn't a crippleware/restricted/dual-licensed scam
to get you hooked: the software is fully-featured, all GPL2.  The
support is CHEAP, at least several tens of thousands of dollars per
year cheaper than the nearest competitor in my situation.

Caveat emptor:

While I can monitor the same number of hosts (though fewer metrics,
less often) with RRD-based systems on a single quad-proc/8GB machine,
Zabbix is now requiring (with ~2000 hosts, ~100,000 values, avg
5-minute resolution) two dedicated 8-way MySQL servers, a 32G SSD and
4TB of FC-SAN.  It doesn't scale the same as RRD's with their
automatic reduction in resolution over time and simple pre-sized
binary format.  It DOES give you the ability to keep as
high-resolution data as long as you want, and amazingly it allows for
actually doing this, which nothing else has even allowed to this
degree of precision and granularity.

Cheers,

/eli



On Jan 18, 2008 2:57 PM, Kris <krisa at subtend.net> wrote:
> Brent Jones wrote:

> > Zabbix:
> > Open source SNMP and Agent-based monitoring tool that performs similar
> > functions as Cacti, but uses installable agents to get more information
> > about a host system, such as running processes, database monitoring, system
> > health (raid status, power supplies, etc).
> > http://www.zabbix.com
>
> I'm going to be trying this out soon.  Currently I'm using a mix of
> nagios and munin for health checks and stats.
>
> Zenoss is another application, but it didn't look as baked as Zabbix
> when I reviewed them a few months ago.