Application Performance Monitoring (APM) as an industry is a seasoned one. Its roots go back to before the emergence of the Internet as a platform. At first confined to the data center, it erupted in response to a need to understand the user experience in the context of the Internet.
The first phase was dominating by passive techniques that attempted to understand the user experience using synthetic transactions. Dominated by vendors who possessed highly distributed points of presence around the Internet, passive monitoring provided site operators with a historical view of how their application performed from an increasingly distributed user base around the globe.
Passive techniques were the first real attempt to measure the user experience. They offered a generic average look at performance largely based on network paths between the point of presence and the application. Because most of the points of presence were located on or near the Internet backbone, they were unable to take into consideration the impact of the "last mile" between the backbone and the client.
Furthermore, the reliance on synthetic transactions often failed to accurately represent a real user interaction or transaction. Thus, the resulting measurements were more indicative of the health of the Internet and availability of an application than that of the user experience.
Passive monitoring systems were very good at identifying intermediate nodes on the Internet as a source of application performance woes. They were not very good at identifying client or application issues nor at monitoring an actual user experience.
These inadequacies gave rise to a more aggressive technique: active monitoring. This phase was dominated by vendors who inserted tiny bits of code into the client application that subsequently provided more accurate performance data. These solutions solved for the last mile as well as recognizing the impact of rendering on application performance. By monitoring live interactions, active monitoring provided a much more realistic perspective on the user experience.
Active monitoring solved for the inadequacies of passive monitoring, and today we can collect data across the entire user experience—from client to network to applications and their back-end systems.
But that doesn't mean we're finished with the evolution of monitoring. Because while we've addressed the basic need to identify existing performance problems, that's no longer good enough. Almost a third (32%) of all customers "would stop doing business with a brand they loved after one bad experience." (PWC)
It isn't enough to identify what went wrong after the fact. It is increasingly important to the business to identify where trouble might occur before it happens.
That’s why we believe the next phase of monitoring will be predictive.
Predictive monitoring is often associated with machine learning for good reason. The ability to predict failure or performance problems is wholly dependent on experience. In the case of application performance, this experience is derived from data about hundreds and thousands of previous user experiences.
With a robust set of data about user experiences that spans the breadth and depth of the entire application data path, analytics can discover patterns and relationships between a complex set of variables such as time of day, location, business function, browser, OS, and network. The ability to predict a problem is based on the relationship and interaction and current state of all these components.
By being able to recognize outlying and anomalous performance of a given component, predictive monitoring can identify the conditions under which user experience might suffer and raise an alert.
If the alert is raised early enough, action can be taken to rectify the situation before it becomes a problem that impacts the business.
We believe that not only is predictive the next evolution of application performance monitoring, but that it is a necessary evolution. As organizations continue to increase reliance on applications, those applications become as critical as any machine on an assembly line to the health of the business.
Just as predictive maintenance is a critical capability in manufacturing durable goods, so will predictive monitoring be a crucial capability in delivering digital experiences.