BCATE: A balanced contention-aware transaction execution model for highly concurrent OLTP systems

Abstract

Web applications like social networking and online shopping are growing rapidly, forcing the OLTP systems to have the ability to efficiently handle numbers of concurrent transactions. Shared-everything models used in conventional OLTP systems, however, face significant problems in concurrency and scalability. With the increment of concurrent threads, the contention among threads increases sharply and degrades the system performance significantly. Shared-nothing models, on the other hand, are ideal for scalability but suffer a lot from data skew problem. In this paper, we propose bCATE, a novel concurrent transaction execution model which divides the database into conflict partitions and detects the conflicts between transactions in each partition. bCATE adopts an efficient thread assignment strategy to alleviate the performance degradation caused by contention and data skew. We conduct extensive empirical studies on our implementation of bCATE on Shore-MT [1] and demonstrate that bCATE can achieve up to 50% performance promotion against other models. © 2013 Springer-Verlag Berlin Heidelberg.