Hybrid Parallelism for XML SAX Parsing

XML has been widely adopted across a wide spectrum of applications. Its parsing efficiency, however, remains a concern, and can be a bottleneck. At the same time, with the trend towards multicore CPUs, parallelization to improve performance has become increasingly relevant. In previous work, we have investigated parallelizing DOM-style parsing and gained significant speedup. For streaming XML applications, however, SAX-style parsing is often required. In this paper, we present a technique and implementation of a parallel XML SAX parser. To handle inherent data dependencies in XML while still allowing reasonable scalability, we use a 4-stage software pipeline with a combination of strictly sequential stages and stages that can be further data-parallelized within the stage. We thus utilize a hybrid between pipelined parallelism and data parallelism. To demonstrate effectiveness, we test this approach on a Linux machine with two Intel Xeon L5320 CPUs for a total of 8 physical cores, and obtain good speedup up to about 8 CPUs.