Optimizing Exploratory Workflows for Embedded Platform Trace Analysis and Its Application to Mobile Devices.

As 5G wireless communication technology is currently deployed, an increasing amount of data is available from mobile devices out in the field. Exploiting this data, also called system traces, recent investigations show the potential to improve the wireless modem design and performance using datacentric approaches. Such data-centric workflows are exploratory and iterative by nature. For instance, time pattern identification is performed by domain experts to derive assumptions on potential optimizations and these assumptions are continuously refined during multiple iterations of data collection, visualization and exploration. In this context, we propose three optimizations to increase the exploration speed in iterative data-centric workflows. First, we present a methodology based on persistent memoization in order to minimize the data processing duration when additional event sequences need to be extracted from a trace. We show that up to 84.5% of the event extraction time can be spared for a typical modem trace data set. Secondly, we present a novel entropy-based data interaction technique for visual exploration of event sequences and finally, a similarity measure to perform subsequence matching in order to assist the user when identifying frequent time patterns in a trace.