VTD-XML Parsing Performance Optimization based on Helper Thread Sampling Prefetching

With the development of cloud computing, big data and AI technologies, the volume of data is expanding rapidly, and the size of XML documents is also increasing. How to improve the processing efficiency of large-scale XML document data has become a new research hotspot. At present, under the multi-core processor platform, the research work on XML parsing and query is mainly focused on parallel parsing and query, and the performance of XML parsing and query has been greatly improved. However, due to the semi-structured characteristic of XML document and the irregular memory access mode, there are a large number of cache misses in the parsing and query programs, and the access delay becomes the bottleneck of the continuous improvement of performance. To solve this problem, this paper proposes a sample-based helper thread prefetching technology. By using idle multi-core resources, the data required by the main thread is predictably prefetched to the last level of shared cache in advance, so as to hide the delay caused by memory access operations, and achieve the goal of optimizing performance. The experimental results show that the cache miss is reduced by more than 80% after helper thread prefetching, and the total XML parsing performance are improved about 10%, the hot function’s performance is improved about 30%.