Lock-free Hash Table on Graphics Processors
Maryam Moazeni, Majid Sarrafzadeh · 2012
Lock-free data structures guarantee higher throughput than lock-based implementations in parallel architectures. This paper presents the first CAS-based lock-free hash table that is based on chaining on GPUs. We achieve an improvement of about 2-8X over lock-free OpenMP implementation. We also achieve over 2-25X speed up over GPU lock-based implementation. We achieve 1.3X to 3.5X increase on throughput with combined Insert and Search workloads over the counterpart OpenMP implementation.