Transactional memory is a novel paradigm introduced to simplify concurrent programming with respect to the traditional lock-based approach. In transactional memory, concurrent threads synchronize via in-memory transactions. A transaction embeds a sequence of shared memory operations whose effect is atomic: if any of the operations takes place, they all do, and if they do, they appear to other threads to take effect instantaneously.

The transactional approach to contention management guarantees atomicity by making sure that whenever two transactions execute conflicting operations to the shared memory, only one of them proceeds.

A major challenge in this approach is to ensure progress and to minimize the overhead caused by the frequent restart of transactions. This requires taking good choices about which of the two conflicting transactions should continue and when and how the other transaction should be restarted.

The talk will introduce the transactional memory paradigm and will discuss the main contention management policies proposed in the literature. Finally, a new contention manager, called Bimodal, will be presented. Bimodal is especially tailored to accommodate read-only transactions without punishing transactions that write most of their duration.