HP Network Node Manager 9 : Discovering and Monitoring Your Network - Examining discovery results (part 4)

State poller

This section describes state poller, which is used in NNMi.

What is state poller?

There are nodes or interfaces in a network, which need more accurate monitoring (more frequent polling). It is not performance effective to set polling intervals for a whole network, based on most frequent polling demand.

How does it work?

State poller allows for the setting of polling intervals and types for classes, interface types, or node types. Configuration is organized by groups, which allows flexibility in maintaining state poller, especially if the network is dynamic and new discoveries need specific polling configuration.

The following screenshot represents the State poller configuration window:

State poller can be configured to use ICMP or SNMP status checks and can be configured for interface or node. This type of configuration is very flexible, but simultaneously complicated, as overlapping may occur. It is OK as long as we know how overlapping works. There are few rules about how NNMi evaluates objects for state polling:

• If the object is interface, then the interface group is evaluated starting from the lowest group number to the highest. The first matched group is applied. No more evaluation is done.
• If the object was not found in the interface group, node groups are evaluated. Again, evaluation is done starting from the group with the lowest group number to the highest. The first matched group is applied. No more evaluation is done. Any interface that didn't match any interface group inherits settings from its node's group settings, which were applied to the node.
• If any object in the device didn't match any state polling rule, default settings are applied.

The following diagram represents the State poller setting evaluation flow:

How to plan state polling

Initially, it may look like a very tricky and difficult task to understand how the NNMi administrator will know what polling settings to set on every single node, and on every interface in a network. Especially when NNMi monitors hundreds, if not thousands of devices, and thousands if not tens of thousands interfaces.

So, forget about the amount of nodes and interfaces. Let's get answers to following questions first:

• Which devices do we plan to monitor?
• What parameters do we want to monitor?
• What maximum delay is tolerated for notifying a state change?

Answers to these questions will give us a pretty clear picture about what the state poller settings should look like. Will we use ICMP or SNMP? Should we care more about traps, or do we plan to poll devices proactively?

As an example, let's take a very generic ISP network that has a backbone network, VIP customers, home users, international links, and peer connections with other ISPs:

It's obvious that not all devices or interfaces have the same demand for polling. VIP customers, international, and peer connections are probably in the top list that ISP most care about in terms of monitoring. These sites will have more frequent polling cycles, while home users will have less frequent polling cycles. So, we already have a rough state polling design. If we analyze each segment more accurately, we could continue increasing a state polling configuration list.

Such assessment as shown in the previous example will be your starting point for decision input about what interface or node groups should be created.

It is good practice to create simple and short groups, which later would be combined into hierarchical levels for monitoring or visualization purposes.

State poller can use ICMP or SNMP queries. Before you decide which one you want to use, it is important to know what each of them does:

• ICMP ping is used to check the availability of each IP address
• SNMP queries nodes for status information, and as SNMP can query data from nodes about more specific parameters, these queries can also be used for fault monitoring or performance data collection (if iSPI for performance is used)

It is hard to estimate how much SNMP traffic will load the network, as polling is designed to use optimized queries. Every time the configuration is saved, state poller recalculates objects, which can be grouped and optimized for common polling groups.

Polling intervals can be set for each group and can be selected from a wide interval range, from as short as one minute to as long as a day or two. Setting polling intervals is pretty tricky, as by setting too short intervals you can heavily decrease the system performance, and by setting too large intervals you will lose accuracy on monitoring.

State poller operations can be checked at any time using the Help | About Network Node Manager i-series table from the main window menu:

The following table describes some presented parameters.

Result	Description
Status	Overall state poller status:
Poll Counters	Collections requested in the last minute Collection completed in the last minute Collections in process
Time to execute skips in the last minute	The number of scheduled polls that did not complete within the polling cycle. Excellent if it's zero. If it's more than zero, then it is recommended to monitor this number and if it keeps increasing, it means that NNMi's poller cannot poll properly and more of the scheduled polls are behind schedule. Recommendation: Check whether all processes are running Fine tune the polling periods and make them less frequent
Stale connections	Are the connections which have not received a response for longer than 10 minutes. Excellent if this number is zero. Otherwise, this parameter needs monitoring and if it keeps increasing, it means that there are problems with polling engine.
Poller result queue length	Excellent if poller result queue length is equal to zero or close to it. Otherwise ovjboss is running out of memory.
State mapper queue length	Excellent if state mapper queue length is equal to zero or close to it. Otherwise, system performance and database should be checked.