Abstract:

Concurrent data structures (CDS) are essential building blocks for parallel programs, and their efficiency significantly affects the overall performance of the programms. Synchronizing concurrent accesses to shared data is a critical challenge for CDS design. Traditionally, lockbased schemes are widelyused. However, lockbased schemes are either errorprone or inefficient. Transactional memory, as a brand new parallel programming paradigm, is proposed. We construct a concurrent linked list upon Intel’s hardware transactional memory and compare its performance with a lockbased and a lockfree counterparts.

Key words: concurrent linked list, hardware transactional memory, practical analysis