Network Management & Monitoring with Linux

In today's world, where all the computing revolves around the concept of networking, the work for system administrators has become more and more overwhelming. It is the mission of maintaining the availability of resources such as routers, hubs, servers and every critical device in the network.

There are many reasons managers would like to monitor network devices: bandwidth utilization, operational state of links, bottlenecks, problems with the cabling or routing information distributed between its devices, etc. Monitoring network activity is also a good starting point for discovering security problems and misbehaviors.

In many cases, the network of an organization includes expensive links to remote networks (WAN) or the Internet, whose costs may be based on traffic volume. It's very important to maintain statistics of traffic going through these links. This is a very common task in Europe, where X.25 links are still very common. These links are charged on the basis of packets transmitted and received.

Other types of links, like Point to Point or Frame Relay, are usually charged on a flat rate. In these, the telco ensures a bandwidth that is important to monitor.

In the final part of this article we focus on a tool designed to monitor traffic in router interfaces, with a great graphical representation of this information. It can be easily modified to monitor other kinds of information.

The answer to all these needs is a protocol named Simple Network Management Protocol (SNMP). Designed in the '80s, SNMP's initial aim was to integrate the management of different types of networks with a simple design that caused very little stress on the network.

SNMP operates at the application level using TCP/IP transport-level protocols so it can ignore the underlying network hardware. This means the management software uses IP, and so can control devices on any connected network—not just those attached to its physical network. This also has disadvantages: if the IP routing is not working correctly between two devices, it's impossible to reach the target to monitor or reconfigure it.

There are two main elements in the SNMP architecture: the agent and the manager. It's a client-server architecture, where the agent is the server and the manager is the client.

The agent is a program running in each of the monitored or managed nodes of the network. It provides an interface to all the items of their configuration. These items are stored in a data structure called a management information base (MIB), which we explain later. It's the server side, as long as it maintains the information being managed and waits for commands from the client.

The manager is the software that runs in the monitoring station of the network, and its role is contacting the different agents running in the network to poll for values of its internal data. It's the client side of the communication.

There is a special command in the SNMP command set called trap that permits an agent to send unsolicited data to the manager, to inform it of events, such as errors, shutdowns, etc.

In essence, SNMP is a very simple protocol as long as all the operations it performs deal with the fetch-and-store paradigm, and this allows for a small commands set. A manager can perform only two different operations on an agent: request or set the value of a variable in the MIB of the agent. These two operations are known as get-request and set-request. There's a command to respond to a get-request called get-response, which is used only by the agent.

The extensibility of the protocol is directly related to the capability of the MIB to store new items. If a manufacturer wants to add some new commands to a device such as a router, he must add the appropriate variables to its database (MIB).

Almost all manufacturers implement versions of SNMP agents in their devices—routers, hubs, operating systems, and so on. Linux is not an exception to this, and publicly available SNMP agents for Linux can be found on the Internet.

SNMP provides very little support for authentication schemes. It supports only a two-password scheme. The public allows managers to request the values of variables, and the private allows these values to be set. These passwords in SNMP are called communities. Every device connected to an SNMP-managed network must have these two communities configured. It is very common to have the public community set to “public” and the private community to “private”, but it's very important to change these values to reflect the security policy of your organization.