A Lightweight Snapshot-based Algorithm for Software Transactional Memory

Abstract

Software transactional memory (STM) is a generic synchronization construct that enables automatic conversion of correct sequential objects into correct nonblocking concurrent objects. We investigate a novel lightweight snapshot-based validation and conflict detection (LSVCD) algorithm for STM that provide scaling and strong consistency. In LSVCD, each transaction maintains a local snapshot version variable used to indicate the snapshot state it views. Whether the snapshot of memory observed by a transaction is consistent or not can be verified quickly by comparing lsv with Global Version Number. LSVCD can greatly reduce the cost of read and write operations, conflict detection and validation of STM. Transactions proceed without interference or contention unless they access overlapping data sets. Experimental results show that LSVCD is simple and quite competitive, and gains a notably improvements of performance over existing STM implementation.