Design And Implementation Of Pallaral Batch B~+-tree Indexing On Multi-core Processors

Recently, microprocessor designs have shifted from fast uniprocessors that exploit instruction level parallelism to chip multiprocessors. It's a greate challenge for software engineer to take advantage of on-chip parallelism by writing applications with a high degree of thread level parallelism. DBMS (Database Management Systems) play an important role in morden life. How to maximize the potential of multicore processors to improve the performance of DBMS becomes an important research topic.This paper focuses on parallel batch B+-tree insertion on multi-core architechtures. Based on this algorithm, we design and implement a storage engine of MySQL called MTPower that supports B+-tree indexing of insert, query, update, and delete. At the same time, we develop a new indexing structure called B bucket, which supports parallel batch insert and query opertatins, to improve the disk sequential access performance.Our tests show that pallaral batch B+-tree insertion performs almost 10 times better than the traditional B+-tree insertion algorithms. For MTPower storage engine with B+-tree index, the performance gain is 4.8 times than MyISAM storage engine. Also, by using 6 threads the performance of B bucket batch insertion can be improved more than 1.6 times than that of a single thread.Our test verifies that using multithreading technique can really improve system performance on multi-core systems. Also it shows that when the total number of threads on multi-core architectures is close to the number of cores of processor the system can achieve the best performance.