A CMDB, or configuration management database, is the single record of what an organization’s IT estate consists of and how its parts depend on each other. Every server, laptop, application, and service is recorded as a configuration item, with the relationships that connect them.

When an incident hits, the first questions are always the same: what is affected, what does it depend on, and who else will notice? A CMDB answers all three, because it holds the map of the estate. Without it, those answers come from memory and guesswork, usually at 2 a.m.

Configuration items and relationships

The building blocks of a CMDB are configuration items, or CIs. A CI is anything the IT team manages: a physical server, a virtual machine, a database, an application, a network switch, or a business service such as email or the company website. Each CI carries attributes — owner, version, location, status, and support details.

What makes a CMDB more than a list is the relationships between CIs. The website depends on a web server, which depends on a database, which runs on a storage array. The email service depends on the mail server and the identity system. Those links are what turn a list of assets into a map of the business.

The most useful relationships are the ones that answer impact questions. If a storage array fails, the CMDB shows every application and service that depends on it. That list is the blast radius of the incident, and it is what the incident team needs in the first five minutes.

How a CMDB gets filled

A CMDB is only useful if it is accurate, and accuracy is the hard part. The two sources of data are discovery and change management, and a healthy CMDB uses both.

Discovery tools scan the network, read installed software, and detect servers, devices, and applications automatically. They find the things nobody remembered to record, including the forgotten server under a desk that turns out to run payroll. Discovery is the only realistic way to keep a large estate current.

Change management is the second source. Every approved change should update the CMDB as part of the change itself: a new server is added when it is built, a retired application is marked retired when it is switched off. A CMDB fed only by manual entry goes stale within months, and a stale CMDB is worse than none, because people act on wrong information.

Note: Discovery finds what exists, but it cannot tell you what something is for. The business context — which service an application supports, who owns it, how critical it is — still has to come from people. Plan for that, or the CMDB will be technically complete and practically useless.

What a CMDB is used for

Incident management is the most common use. When a server fails, the CMDB shows what depends on it, so the team knows the impact before users start calling. It also shows the CI’s owner, its support contract, and its recent change history, which is often where the cause hides.

Change management is the second use. Before approving a change, the change board checks the CMDB for what the change touches and what depends on it. A patch to a database server looks harmless until the CMDB shows it also hosts the billing system.

The third use is planning. A CMDB shows which systems are out of support, which have no owner, and which are duplicated. That is the evidence for consolidation, upgrades, and license renewals — and for saying no to the next purchase of something the organization already owns.

Common mistakes

The classic mistake is trying to record everything. A CMDB that tracks every cable and every minor setting becomes so expensive to maintain that nobody maintains it. The right scope is the items and relationships that real decisions depend on, and nothing more.

The second mistake is treating the CMDB as a project with an end date. It is not something you build once and finish. It is a living record that needs continuous feeding from discovery and change, and a named owner who is accountable for its accuracy.

The third mistake is buying the tool before defining the scope. The CMDB product matters less than the answer to two questions: which CIs will we track, and who keeps them accurate? Answer those first, and the tool choice becomes much easier.

Try the tool

Reliability starts with knowing what you have. When a system fails, the question is how long it runs between failures and how fast you recover — and both numbers come from the CMDB’s history. Use the MTBF calculator to work out the average time between failures for a system, and compare it against the systems the CMDB says it depends on.