CDC Cyber Defence Center Recovery Security Root Cause Analysis
Home Securiy Playbooks Security Disaster Recovery Security Root Cause Analysis
In science and engineering, root cause analysis (RCA) is a method of problem solving used for identifying the root causes of faults or problems. It is widely used in IT operations, telecommunications, industrial process control, accident analysis (e.g., in aviation, rail transport, or nuclear plants), medicine (for medical diagnosis), healthcare industry (e.g., for epidemiology), etc.
RCA can be decomposed into four steps:
Identify and describe the Problem clearly.
Establish a timeline from the normal situation up to the time the problem occurred.
Distinguish between the root cause and other causal factors (e.g., using event correlation).
Establish a causal graph between the root cause and the problem.
RCA generally serves as input to a remediation process whereby corrective actions are taken to prevent the problem from reoccurring. The name of this process varies from one application domain to another.
In science and engineering, there are essentially two ways of repairing faults and solving problems.
Reactive management consists in reacting quickly after the problem occurs, by treating the symptoms. This type of management is implemented by reactive systems, self-adaptive systems, self-organized systems, and complex adaptive systems. The goal here is to react quickly and alleviate the effects of the problem as soon as possible.
Proactive management, conversely, consists in preventing problems from occurring. Many techniques can be used for this purpose, ranging from good practices in design to analyzing in detail problems that have already occurred, and taking actions to make sure they never reoccur. Speed is not as important here as the accuracy and precision of the diagnosis. The focus is on addressing the real cause of the problem rather than its effects.
Root-cause analysis is often used in proactive management to identify the root cause of a problem, that is, the factor that was the main cause of that problem.
It is customary to refer to the root cause in singular form, but one or several factors may in fact constitute the root cause(s) of the problem under study.
A factor is considered the root cause of a problem if removing it prevents the problem from recurring. A causal factor, conversely, is one that affects an event's outcome, but is not the root cause. Although removing a causal factor can benefit an outcome, it does not prevent its recurrence with certainty.
Imagine an investigation into a machine that stopped because it overloaded and the fuse blew. Investigation shows that the machine overloaded because it had a bearing that wasn't being sufficiently lubricated. The investigation proceeds further and finds that the automatic lubrication mechanism had a pump which was not pumping sufficiently, hence the lack of lubrication. Investigation of the pump shows that it has a worn shaft. Investigation of why the shaft was worn discovers that there isn't an adequate mechanism to prevent metal scrap getting into the pump. This enabled scrap to get into the pump, and damage it.
The apparent root cause of the problem is therefore that metal scrap can contaminate the lubrication system. Fixing this problem ought to prevent the whole sequence of events recurring. The real root cause could be a design issue if there is no filter to prevent the metal scrap getting into the system. Or if it has a filter that was blocked due to lack of routine inspection, then the real root cause is a maintenance issue.
Compare this with an investigation that does not find the root cause: replacing the fuse, the bearing, or the lubrication pump will probably allow the machine to go back into operation for a while. But there is a risk that the problem will simply recur, until the root cause is dealt with.
Root-cause analysis is used in many application domains.
Manufacturing and industrial process control
The example above illustrates how RCA can be used in manufacturing. RCA is also routinely used in industrial process control, e.g. to control the production of chemicals (quality control).
RCA is also used for failure analysis in engineering and maintenance.
IT and telecommunications
Root-cause analysis is frequently used in IT and telecommunications to detect the root causes of serious problems. For example, in the ITIL service management framework, the goal of incident management is to resume a faulty IT service as soon as possible (reactive management), whereas problem management deals with solving recurring problems for good by addressing their root causes (proactive management).
Another example is the computer security incident management process, where root-cause analysis is often used to investigate security breaches.
RCA is also used in conjunction with business activity monitoring and complex event processing to analyze faults in business processes.
Health and safety
In the domains of health and safety, RCA is routinely used in medicine (diagnosis), epidemiology (e.g., to identify the source of an infectious disease), environmental science (e.g., to analyze environmental disasters), accident analysis (aviation and rail industry), and occupational safety and health.
RCA is also used in change management, risk management, and systems analysis.
Example of a root cause analysis method
Despite the different approaches among the various schools of root cause analysis and the specifics of each application domain, RCA generally follows the same four steps:
Identification and description: Effective problem statements
and event descriptions (as failures, for example) are helpful and usually
required to ensure the execution of appropriate root cause analyses.
Chronology: RCA should establish a sequence of events or timeline for understanding the relationships between contributory (causal) factors, the root cause, and the problem under investigation.
Differentiation: By correlating this sequence of events with the nature, the magnitude, the location, and the timing of the problem, and possibly also with a library of previously analyzed problems, RCA should enable the investigator(s) to distinguish between the root cause, causal factors, and non-causal factors. One way to trace down root causes consists in using hierarchical clustering and data-mining solutions (such as graph-theory-based data mining). Another consists in comparing the situation under investigation with past situations stored in case libraries, using case-based reasoning tools.
Causal graphing: Finally, the investigator should be able to extract from the sequences of events a subsequence of key events that explain the problem, and convert it into a causal graph.
To be effective, root cause analysis must be performed systematically. A team effort is typically required. For aircraft accident analyses, for example, the conclusions of the investigation and the root causes that are identified must be backed up by documented evidence.
Transition to corrective actions
The goal of RCA is to identify the root cause of the problem. The next step is to trigger long-term corrective actions to address the root cause identified during RCA, and make sure that the problem does not resurface. Correcting a problem is not formally part of RCA, however; these are different steps in a problem-solving process known as fault management in IT and telecommunications, repair in engineering, remediation in aviation, environmental remediation in ecology, therapy in medicine, etc.
Without delving in the idiosyncrasies of specific problems, several general conditions can make RCA more difficult than it may appear at first sight.
First, important information is often missing because it is generally not possible, in practice, to monitor everything and store all monitoring data for a long time.
Second, gathering data and evidence, and classifying them along a timeline of events to the final problem, can be nontrivial. In telecommunications, for instance, distributed monitoring systems typically manage between a million and a billion events per day. Finding a few relevant events in such a mass of irrelevant events is akin to finding the proverbial needle in a haystack.
Third, there may be more than one root cause for a given problem, and this multiplicity can make the causal graph very difficult to establish.
Fourth, causal graphs often have many levels, and root-cause analysis terminates at a level that is "root" to the eyes of the investigator. Looking again at the example above in industrial process control, a deeper investigation could reveal that the maintenance procedures at the plant included periodic inspection of the lubrication subsystem every two years, while the current lubrication subsystem vendor's product specified a 6-month period. Switching vendors may have been due to management's desire to save money, and a failure to consult with engineering staff on the implication of the change on maintenance procedures. Thus, while the "root cause" shown above may have prevented the quoted recurrence, it would not have prevented other – perhaps more severe – failures affecting other machines.