Profilers help developers to find and fix performance problems. But do they find performance bugs -- performance problems that real users actually notice? In this paper we argue that -- especially in the case of interactive applications -- traditional profilers find irrelevant problems but fail to find relevant bugs.

Many computer systems are interactive in some way, that means they are used by human users. A human user perceives the performance of a computer system primarily in terms of its response time. If a system does not respond to a user's input, such as a key press, a mouse motion, or a gesture on a touch screen, within roughly 100 ms, a user perceives the system as sluggish. Developers of interactive software and systems thus are interested in keeping response times below this perceptibility threshold. Existing tools allow the measurement of interactive response times, and the tracing of application behavior. In this paper we present a tool, LagAlyzer, which analyzes and visualizes the information gathered by such latency measurement tools. LagAlyzer enables the characterization of perceptible lag, for example by quantifying to what degree perceptible performance was caused by synchronization bottlenecks, by garbage collection, by the runtime libraries, or by the application. We use LagAlyzer to characterize the perceptible latency found in commonly used interactive Java applications. We believe that this is the first study giving insight into why interactive Java applications sometimes are perceived as sluggish.