Milan Jovic

Understanding the Performance of Interactive Applications Milan Jovic

Many if not most computer systems are used by human users. The performance of such interactive systems ultimately affects those users. Thus, when measuring, understanding, and improving system performance, it makes sense to consider the human user's perspective. Essentially, the performance of interactive applications is determined by the perceptible lag in handling user requests. So, when characterizing the runtime of an interactive application we need a new approach that focuses on the perceptible lags rather than on overall and general performance characteristics. Such a new characterization approach should enable a new way to profile and improve the performance of interactive applications. Imagine a way that would seek out these perceptible lags and then investigate the causes of these lags. Performance analysts could simply optimize responsible parts of the software, thus eliminating perceptible lag for interactive applications. Unfortunately, existing profiling approaches either incur significant overhead that makes them impractical for an interactive scenario, or they lack the ability to provide insight into the causes of long latencies. An effective approach for interactive applications has to fulfill several requirements such as an accurate view of the causes of performance problems and insignificant perturbation of the interactive application. We propose a new profiling approach that helps developers to understand and improve the perceptible performance of interactive applications and satisfies the above needs.