CS department's Talk Series, Winter Term 2015/16
, TU Kaiserslautern
Kaiserslautern, Germany Tue 10 November 2015
Destructible updates on shared objects require careful handling of concurrent accesses in multi-threaded programs. Paradigms such as Transactional Memory support the programmer in correctly synchronizing access to mutable shared data by serializing the transactional reads and writes. But under high contention, serializable transactions incur frequent aborts and limit parallelism. This can lead to a severe performance degradation. We propose mergeable transactions which provide a consistency semantics that allows for more scalability even under contention. Instead of aborting and re-executing, object versions from conflicting updates on shared objects are merged using data-type specific semantics. The evaluation of our prototype implementation in Haskell shows that mergeable transactions outperform serializable transactions even under low contention while providing a structured and type-safe interface.
Keywords : Scalable Consistency, Concurrent Data Types, Parallel ComputingDetails Presentation