Parsing XML Using Parallel Traversal of Streaming Trees

Abstract

Abstract. XML has been widely adopted across a wide spectrum of applications. Its parsing efficiency, however, remains a concern, and can be a bottleneck. With the current trend towards multicore CPUs, parallelization to improve performance is increasingly relevant. In many applications, the XML is streamed from the network, and thus the complete XML document is never in memory at any single moment in time. Parallel parsing of such a stream can be equated to parallel depth-first traversal of a streaming tree. Existing research on parallel tree traversal has assumed the entire tree was available in-memory, and thus cannot be directly applied. In this paper we investigate parallel, SAX-style parsing of XML via a parallel, depth-first traversal of the streaming document. We show good scalability up to about 6 cores on a Linux platform.

A 1 cycle-per-byte XML parsing accelerator

“…A large array of research results have been reported, which often exploit the SIMD instruction set extension of CPUs, or multicore CPUs to speed up XML processing in software [9][13] [14] [16]. However, their results are often incomplete, e.g.…”