2PLSF: Two-Phase Locking with Starvation-Freedom.

Invented more than 40 years ago, the two-phase locking concurrency control (2PL) is capable of providing opaque transactions over multiple records. However, classic 2PL can suffer from live-lock and its scalability is low when applied to workloads with a high number of non-disjoint read accesses. In this paper we propose a new 2PL algorithm (2PLSF) which, by utilizing a novel reader-writer lock, provides highly scalable transactions with starvation-freedom guarantees. Our 2PLSF concurrency control can be applied to records, metadata and to indexing data structures used in database management systems (DBMS). In our experiments we show that 2PLSF is often superior to classical 2PL and can surpass concurrency controls with optimistic reads, simultaneously providing high throughput and low latency for disjoint and non-disjoint workloads.