Enhanced Encoding Techniques for the Open Trace Format 2

AbstractHighly effcient encoding of event trace data is a quality feature of any event trace format. It not only enables measurements of long running applications but also reduces bias caused by intermediate memory buffer flushes. In this paper we present encoding techniques that will remarkably increase memory effciency without introducing overhead for the compression. We applied these techniques to the Open Trace Format 2, a state-of-the-art Open Source event trace data format and library used by the performance analysis tools Vampir, Scalasca, and Tau. In addition, we show that these encoding techniques are a basic step in achieving a complete in-memory event trace work flow