[AusNOG] Network Management and Tools

[Andrew Fort]

On Thu, Jul 8, 2010 at 1:56 AM, phil colbourn <philcolbourn at gmail.com> wrote:
> Again, that you for sharing your use of network management tools.
> I have roughly tallied the main products mentioned: (I counted OpsView as
> Nagios+MRTG)
> Nagios  7
> MRTG  3
> Cacti  4
> RANCID  5
> and notably
> PRTG  2
> Collectd  2

I don't know why collectd, something which does very little, fits the
unix pipeline mentality, and is thus incredibly useful for sysadmins
and toolwriters, is "notable", but it's pretty good :).

> Many use custom in-house-developed databases, scripts and tools.

The "every network is different" bit applies here.  Turns out most of
the scripts people use that are custom are tying together their
pipelines.

Because your network (or gear) may be service oriented, your database
probably has customers and services in it, too; that's fine, but it's
going to differ from a content provider who has only peers and mostly
runs a SP BGP kinda network.

And finally because you'll come at things from different levels of
requirements, maturity, time/money budgets.  Letting protocols deal
with most of the issue without them being modelled seems generally
fine.  At some point you're going to need to automate some knobs, at
which point you do the extra work.

> A significant number use commercial products.
> Most described alarm/event/performance management or configuration backup
> products
> A few mentioned configuration generation systems.

No-one mentioned how to actually deploy that stuff to routers.  What
if I have to deploy to 10 different hardware platforms and
so-on.  Are people just wrapping RANCID scripts for this?   (I used to
do it this way, quite handy).  At some point years ago I needed
something a little more complicated and so had the idea for an agent
to manage these CLI connections.  I wrote that
(http://code.google.com/p/notch) open source after having written it
at my last employer.  But mostly it's just been my itch to scratch.

What other tools are people using for deployment of configuration to
routers?  I mean everything from config + bootp/dhcp to in-place
automation.  I'm less interested in the "we use 5620 SAM" (Alcatel's
sales droids don't have to work hard here - it really doesn't suck!),
but more interested in the "here's how we do that for 5 different
vendors' gear".  That's the problem I've been interested in solving
with open source software, and IMO it's better than the commercial
approaches here.

> Does that seem about right? Are these typical of what network operators are
> using?

In Australia, yes.  An exhaustive list?  Certainly not.

> No one mentioned (other NMS products that I have heard of) HP OpenView,
> Tivolli, NetCool, cricket, OpenNMS, ZenOSS or even SNMP2XML ;) ?

Plenty of people use OpenNMS and ZenOSS.  I've a little experience
with ZenOSS and its Python is not a good posterboy for the language
(good Perl == Radiator.  good python == most anything except zenoss,
it seems).  It's relatively powerful, but I dislike the "we do
everything" approach, personally.

>
> What concerns me is that once upon a time we used Nagios/NetSaint,
> big-brother, cricket, MRTG, and looked at Rancid, Cacti, OpenNMS and
> ZenOSS.
> Have we moved in the wrong direction?

What do you mean; people are apparently using less "integrated" NMSes,
so this is a bad thing?  If so, I disagree.

Big NMSes, million dollar software with $3000/day consultants, should
this be encouraged?  I think not.  Instead, start from the ground up,
with a few basic requirements.  You don't need to spend big money to
do that.  Conversely, a million dollar price tag doesn't mean they
have customers who use the software to do what you do. Shocking, I
know.  The important part is that you need your best operators,
architects and engineers to talk about this long enough to get a set
of requirements and a design isolated.

Again referencing Stephen Stuart's NANOG26 presentation about these
things: "perfect is the enemy of done.".

-a