Composability for Application-Specific Transactional Optimizations

Software Transactional Memory (STM) has made great advances towards acceptance into mainstream programming by promising a programming model that significantly reduces the complexity of writing concurrent programs. Unfortunately, the mechanisms in current STM implementations that enforce the fundamental properties of transactions — atomicity, consistency, and isolation — also introduce considerable performance overhead. This performance impact can be so significant that in practice, programmers are tempted to leverage their knowledge of a specific application to carefully bypass STM calls and instead access shared memory directly. While this technique can be very effective in improving performance, it breaks the consistency and isolation properties of transactions, which have to be handled manually by the programmer for a specific application. It also breaks another desirable property of transactions: composability. In this paper, we identify the composability problem and propose two STM system extensions to provide transaction composability in the presence of direct shared memory reads by transactions. TxFastRead gives the programmer the performance of a direct memory access to the shared memory data when used from a non-nested transaction, while performing necessary bookkeeping to guarantee composability when used from a nested transaction. TxFlush ensures consistency between nested transactions that use TxFastRead and can be inserted by the programmer or automatically by the compiler. These extensions give the programmer a similar level of flexibility and performance when optimizing the STM application as existing practices, while preserving composability. We propose two implementation schemes for these extensions: Lookup Scheme and Partial Commit Scheme. We evaluate our implementation of these extensions on several benchmarks on a 16-way SMP. The results show that our extensions provide performance competitive with hand-optimized non-composable techniques, while still maintaining transactional composability.