Hardware Accelerated Parallel Speculative XML Parser

The extensible markup language XML is a standard information representation tool and playing an increasing important role in many fields, like database and web services. XML parsing is a core task and has great influence on XML applications'performance. Almost all of the commercial XML parsers are software implemented and run in serial. In order to accelerate XML parsing, parallel XML parsing method is introduced. Currently, traditional processors can not meet the needs of software applications, so use hardware as a coprocessor can increase performance. Some researchers presented that an FPGA implemented XML parser, and it parses XML document in sequence. However, it does not make full use of hardware's highly parallel advantage, thus it increases limited performance.In this paper, a parallel speculative XML parsing method is presented, and it can accelerate XML parsing with fine granularity parallelism. A sub-XML document partition algorithm is presented which can divide a XML document into several sub documents and each sub document will be parsed in parallel. Under the situation that the correctness of XML document's well-formed checking and schema validation is unknown, a speculative DOM-tree is built in memory in order to increase performance.The parallel speculative Dom-based XML parser (PSDXP) includes parallel well-formed checking, parallel DOM-tree construct and parallel schema validation. Our PSDXP runs on 125 MHZ frequency. The PSDXP is implemented of FPGA both with two processes parallelism and four processes parallelism on a Xilinx Virtex-5 board. The hardware resource usage of PSDXPx2 is 5% Slice Register, 9% Slice LUT and 8% Block RAM. It has 2.657 times Slice Register and 2.639 times than XPA. The hardware resource usage of PSDXPx4 is 11% Slice Register, 19% Slice LUT and 17% Block RAM. It has 6.416 times Slice Register and 5.895 times than XPA.The PSDXP is compared with other two XML parsers, one is software implemented (RapidXML) and the other one is FPGA implemented (XPA). PSDXP and XPA run on 125 MHZ, RapidXML runs on 3.16 GHZ. Experimental results show that PSDXPx4 has 1.342 times better than RapidXML in metric of Gbps and 36 times in metric of CPB. And PSDXPx4 has 4 times better than XPA both in metrics of Gbps and CPB. Thus, the parallel speculative XML parsing architecture this paper proposed is better than other XML parsers in performance, and the resource usage is not so much, also with good scalability.