Technical Report Number
Instrumentation and log data monitoring are important features present in many different applications today that are used in analysis of a system behaviour. They are largely used in collection of key information from the system which might allow us to describe or establish certain facts about the behavior of the system. Instrumentation in any application usually comes along with its overheads. In some cases , code for instrumentation might become more expensive on the underlying resources than the application itself. For many applications, especially real-time applications , this overhead can cause serious and often catastrophic system failures. Collecting useful information from a real-time application has to be without interfering with the performance of the application. In this paper , I describe a mechanism of collection and visualization of performance data from a real-time concurrency platform in linux , without interfering and ensuring the correctness of running real-time parallel tests and experiments. This paper describes two different forms of data collection 1) Collection of timing related information from various critical phases in a real-time experiment , 2) Collection of cpu preemption information on running experiment threads. Data collected from both these stages are further fed to a visualization tool for post-debugging analysis of real-time experiment.
Mahadevan, Mahesh, "Data collection and performance monitoring of real-time parallel systems " Report Number: WUCSE-2012-76 (2012). All Computer Science and Engineering Research.
Permanent URL: http://dx.doi.org/10.7936/K7RX999J