[c-nsp] Free NMS Tools

[Saku Ytti]

On (2009-07-03 14:00 +0100), Mario Spinthiras wrote:

Hey,

> I would say Zenoss is looking good because of the inventory management you
> can do and because of the logical structure it puts everything in. I wrote
> 
> Everything else just seems inadequate or poor.

I recently spent few moments evaluating zenoss and was not impressed. To me
all OSS NMS solutions out seem like they are made by coder-in-server-admin
not coder-in-network-admin, and as such seem to have much more integration
with servers than with network, zenoss seems like no exception.

My main grief with NMS I've looked at is virtually no integration with network
devices out of the box. Why don't they ship with MIBs or just specific OIDs
for few top vendors important traps etc? Adding appropriate reaction
classification. Networking is comparatively homogeneous environment, unlike
server admins who have high variance in OS and applications, network
operators out there have very similar requirements, allowing very advanced
integration out-of-the-box. People want NMS to automatically monitor BGP,
OSPF, IS-IS, LDP, status of some other CPU/memory than just control-plane
pending few minutes thinking it would be easy to add lot of really common
things here, that would be desired by very many network operators.

Other thing that annoys me is how SNMP pollers are implemented, they're
blocking, giving sucky performance on misbehaving or down nodes. And
even still puzzlingly slow. While having SNMP poller poll 140k OID
per second on 386 class PC is rather trivial, using two process strategy,
where single process spews packets outs, and another listens what comes
back, completely asynchronous, agnostic to any problem host may have.
I've also only seen alarms based on absolute values of different counters,
like CPU, memory, iface error counters etc. While I'd like automatic
trending alarms, so if my memory use for past 5 months was relatively
static, then for few consecutive days has increased steadily, it is
likely memory leak, and I want to know about it, even if I have GB's of
free memory. This type of 'trending' module should be relatively 
easy, and could be reused by any counter values.

I demoed zenoss with 27 routers and it froze trying to poll their
interface (granted there are very many interfaces). (2.3GHz Intel,
with 2GB of memory), turning performance graphs off helped, of course.
Trying to use zenpacks to add (3rd party provided) Cisco MIBs took
hours and failed due to exhausted disk space, not sure which device
it was, as it didn't tell, but smallest is /tmp with 186MB free.

I'd be happy to pay zenoss enteprise costs, if it would have
basics integration with network, but value it actually delivers
to me, is actually so modest, you can pick up any other 
NMS there or hack something on your own. Since most time would
be committed anyhow adding basic functionality.

Thanks,
-- 
  ++ytti