"Enterprise" level features

[Alexey_Busov]

Good day,

as I work with "real" enterprise-level (it's pice in M$)network performance trending and monitoring s/w may I suggest here some strategic implovements :

1/ there shold be 4 layers in the system
a/ discovery and creation of "instances" to be monitored
b/ collection of KPIs for the instance + abstraction + aggregation over time
c/ abstracted data processing for instances across topology (for ex. max util over group of servers)
d/ presentation

2/ "instances" monitored by snmp should be at least
a/ hosts (i.e. switch chassis, server) identified by IP, node type
b/ interfaces (i.e. Ethernet0/1 ) identified by IP of the node it belongs to, ifIndex(ifDescr), ifType, etc.

3/ Instances should have "properties" configurable automatically or manually

4/ as monitored networks in Enterprises go for thousands of nodes, focus in development IMHO should be on automation of all processes - from discovery of new topology to presentation of reports to new user(s).


Alexey

[nelsonab]

While your suggestions are valid, I'm sure there are people who may not agree that they are necessary, or that they are overall very important. However Zabbix is driven by it's user base. I hope that this would not be your last post here but rather that you begin a continuing relationship with the Zabbix community. Things are getting better and I invite you to be part of those improvements.

[fmrapid]

Zabbix meta data

I concur with much you are saying.

3/ See the following feature request which deals with exactly what you are saying. You can and others can vote for it.

https://support.zabbix.com/browse/ZBXNEXT-577

4/ Low level discovery and actions associated with discovery in version 2.0 seems to be a step in this direction. I am not sure how much it does implement to feed point 3/ and how flexible it is in regards to monitoring network devices.

fmrapid

[Alexey_Busov]

Discovery via "topology" import

Just recalled one more feature that would greatly improve automation of monitoring:

Discovery should be able to process "topology" definition files (simplest form is ip list in txt file)
More complex form would define "instances" i.e. nodes(chassis) and interfaces to be monitored.

just think of a task to monitor 2.1M cable modems we are facing. Ip's are dynamic and not from continious IP ragnes. Even split into regions (hubs) of ~10k there is no way to configure it in Zabbix.

This is a very common "Enterprise" situations.

Cheers,
Alexey

[qix]

One of the things I really just love about Zabbix is it's flexibility.
I'm sure that with a bit of scripting/api work you will get your 2.1M nodes in Zabbix in no time.

[untergeek]

I work in hardware device management (Wireless, DSL, Fiber, etc). Zabbix is well-suited to monitoring the application servers that do the hardware device management. It's not very well-suited to doing the hardware device management itself.

2.1M cable modems is not a trivial number, and there's a customer attached to each device. I'd be curious to see if you can get Zabbix to do what it seems you're trying to do with it.

[Alexey_Busov]

I totally agree with untergeek that Zabbix is not very much geared towards h/w or appliances monitoring via SNMP (for example)

And that is why I have started this thread.
This is my input to make this s/w more usable for people monitoring networks.

While I agree with "qix" that with help of scripting and API everything can be done, I have enough load from monitoring itself, so I'd prefer to have all features in the GUI.
Also "Enterprise" level monitoring system should not have that much of scripting required.

if we find Zabbix able to poll 15 data points every 15 min from 2.1M modems I'll post results here. I plan to start working on this in a couple of months.

Cheers,
Alexey