Evaluating the Performance of CSB+-Trees on Multithreaded Architectures

Modern architectures have made considerable increases in processor speed and performance. However, database management systems (DBMSs) fall far short from achieving their ideal performance. DBMSs are widely used in almost every large organization. Therefore, it is important to achieve fast data retrieval and processing. Recent studies have shown that more than 50% of the execution time in database operations is spent waiting for data. CSB+-trees were introduced to speedup index structure operations, mainly the search and update. In this paper we propose a multithreading technique to utilize the two threads available in an Intel Pentium 4 Hyperthreaded (HT) platform. Our technique gains speedup ranging from 29% to 70% for dual-threaded CSB+-tree on an HT enabled platform compared to a single-thread version running on HT disabled architecture.