Why Transactional Memory Should Not Be Obstruction-Free

Transactional memory (TM) is an inherently optimistic abstraction: it allows concurrentprocesses to execute sequences of shared-data accesses (transactions) speculatively, with anoption of aborting them in the future. Early TM designs avoided using locks and relied onnon-blocking synchronization to ensure obstruction-freedom: a transaction that encounters nostep contention is not allowed to abort. However, it was later observed that obstruction-free TMs perform poorly and, as a result, state-of-the-art TM implementations are nowadaysblocking, allowing aborts because of data conflicts rather than step contention.In this paper, we explain this shift in the TM practice theoretically, via complexity bounds.We prove a few important lower bounds on obstruction-free TMs. Then we present a lock-based TM implementation that beats all of these lower bounds. In sum, our results exhibit aconsiderable complexity gap between non-blocking and blocking TM implementations.