System-level Monitoring

A second approach to application performance management is to measure system-level parameters like CPU usage, memory utilization, number of open files or run state of a process or a thread. By mapping processes and threads to applications, status information about the application is gained. The main advantage of this approach is the great experience gathered over the last years which allows easy collection of the data. It is relatively easy to read this kind of information from the system and provide it to management systems via well-defined interfaces.

However, there are major drawbacks of this technique as well: The basic problem of this solution is that it cannot provide the QoS parameters agreed with customers. Information about the state of the underlying systems is of minor importance to technically not versed customers. Mapping of system-level information to user-oriented parameters is often impossible. Even if a process is running perfectly from a systems view, the transactions a user is interested in might still be failing. Therefore, the monitoring of system-level parameters can be invaluable for a provider to monitor overall system performance but cannot be used for measuring actual application performance and verifying SLAs.

The IETF makes heavy use of this approach by providing a number of MIBs (e.g., SysAppl MIB [#!rfc2287!#]) concerning the area of application management. A lot of management tools, like HP Perfview [#!hp01!#] also follow this approach.