A Hybrid TM for Haskell ∗

Much of the success of Haskell’s Software Transactional Memory (STM) can be attributed to the language’s type system and purity, which together allow transactions to be expressed safely and precisely. By construction, transactions are free from many of the perils that other systems must work hard to avoid. Users do have to work in a constrained environment as a result, but the popularity of Haskell’s STM indicates that this is not a burden too hard to bear. At the same time, the performance of Haskell’s STM does not reach the level achieved by recent systems for other languages. The use of Hardware Transactional Memory (HTM) is just beginning, with Intel’s release of the first widely available TMcapable processors. There has been excellent recent work building hybrid transactional memory systems that combine the performance benefits of HTM and the progress guarantees of STM. In this paper we present our ongoing work to build a hybrid transactional memory system for Haskell. Haskell’s restricted environment provides an opportunity for us to explore designs that leverage compile-time information to improve performance while preserving safety. At the same time, some of the features of Haskell STM prove to be challenging to support efficiently