1. Overview

EMT exists to bridge a gap between system and application-specific performance metrics gathering. It provides an easy way to gather common system performance metrics, as well as providing a simple plugin-based interface to collect custom application-specific metrics. The types of systems EMT works well with fit into one of three types:

• Systems with no historical performance monitoring. When something happens to the applications on these systems, someone will login, run top or vmstat and try to diagnose the issue. They have almost no historical information to look for trends in.
• Systems with standard system monitoring provided by sar or something like Zabbix. This is a good start, but it gives you no way to compare the performance of applications running on a system to the performance of the system itself.
• Systems with system stats and some method of gathering application-specific stats. The application stats either aren't historical or are kept completely separate from system stats, so there is no way to correlate application performance problems with the systems the application runs on.

Since EMT is plugin-based, it can collect system stats and correlate them with application-specific stats. This is particularly important when using a tiered approach to monitoring, where each layer in a system monitors the performance of the layer it depends on. For example, if you were monitoring a simple LAMP stack, the webservers would track the system stats, such as CPU, disk busy, and network I/O. It would also collect response time from Apache. EMT would be installed on the databases servers to collect MySQL stats.

If response time on the webservers started to increase, first the system stats on the webserver should be compared to response time. If the system stats are OK, move down the stack to the databases. There will likely be queries taking a long time to return or some other similar issue. This is a fairly simple example but the tiered monitoring approach becomes more and more valuable with increased numbers of layers, especially in systems based on SaaS architecture.