Gallery

Open source server monitoring

Posted on 20 Aug 2009 at 13:32

Simon Brock investigates the best tools for monitoring your server's performance

Computers get faster and smaller every year, but in the case of servers – the building blocks for many modern businesses – the tasks we expect them to perform have increased to match. We so rely on these servers that we increasingly need to monitor what they do, how they do it and when they hit problems.

An underappreciated aspect of modern server designs is that they now incorporate facilities to monitor themselves and report problems: a small computer built into a bigger server to monitor it used to be reserved for very expensive systems, but now even the cheapest 1U Dell server comes with this kind of monitoring and control.

The trouble is that getting information out of these devices is often more difficult than it ought to be. Most of the monitors themselves are built to the same standard, but the tools to take advantage of them tend to be vendor specific, and the formats for accessing the information often impenetrable. So this month I’m going to look at monitoring facilities and open-source solutions for extracting the information they provide.

IPMI one more time

In principle we should check the health of all our hardware through something called Intelligent Platform Management Interface or IPMI, a facility that's built into most modern server motherboards (and quite few desktops too). This defines a set of resources and a set of protocols that allow you to monitor and control the computer. Any machine that supports IPMI will offer the facility in two ways, locally through some form of device driver and remotely via a network interface.

Please heed this security warning: if you enable IPMI on your machine (and be warned that it may already be enabled) then someone else could take over your computer

Before I go into more detail about IPMI, please heed this security warning: if you enable IPMI on your machine (and be warned that it may already be enabled) then someone else could take over your computer.

A user with access to the IPMI facility can power cycle your server and gain access to control functions like console serial ports. On both Unix and Windows machines the serial console allows a user to carry out administrator tasks: for example Windows Server 2003 offers something called the System Administrator Console (SAC) via a serial port by default, which needs no authentication and gives you administrator access to most aspects of the machine. IPMI should be locked down and treated with as much caution as an actual administrator account, and if it's enabled over the network a remote user could power cycle the machine.

So how do you get started with IPMI? Normally IPMI presents itself in two out of these three ways:

1 Your operating system will offer a local device driver that enables access to the IPMI facility built into your machine

2 IPMI will be available over the network in some way. For many servers IPMI is built into the motherboard and implemented by a component called the Baseboard Management Controller (BMC). In Dell machines, for example, this is attached to one of the built-in Ethernet ports and at boot time you're invited to configure it via a BIOS prompt.

Note that while IPMI shares an Ethernet port, it doesn't interfere with any other service running on that port. When you configure IPMI over Ethernet it will require its own IP address and use its own Ethernet address. IPMI controllers like these don’t typically respond to pings or non-IPMI traffic, so make sure you have the port set up correctly and an IPMI client to talk to it.