Not all network management solutions are alike, even though they may sound that way sometimes. A sure sign of a network management tool trying to pass itself off as an enterprise solution is when it implements distributed polling as a way to scale.
These systems scale server capacity by adding distributed pollers, each sharing a portion of the overall CPU load. But any network monitoring architect can tell you that the bottleneck in infrastructure management is I/O to the database. Having multiple pollers simultaneously send data back to a single data store does not solve the issue but can exacerbate it.
This is why network management systems that claim to scale using distributed polling engines can only achieve small increases for each engine—typically only 7,000 to 10,000 additional objects each versus up to 70,000 of an intelligently architected enterprise-capable system. Metaphorically speaking, distributed pollers do allow a network management application to pour more water in the top of the funnel, but the neck of the funnel is the problem.
Other risks include:
- Single point of failure—If the central database fails, the ability of polling servers to collect data will be impacted.
- WAN link failure—A failure of a WAN link between remote pollers and the central data store will cause loss of data.
- Expensive WAN links—If data is sent to the central server over expensive and/or low capacity WAN links then pricy upgrades to these links may be needed.
- Lack of real-time data—If the remote pollers simply gather data and forward it without real-time analysis, the benefits of immediate notification of anomalies are lost.
A true enterprise class solution distributes not only the polling but also the data storage. These multi-server solutions allow each server the visibility to the data stores of the other servers and therefore offer near limitless scalability. This is an architecture designed from its inception for enterprise computing.