The foreseen evolution of chip architectures tohigher number of, heterogeneous, cores, with non-uniformmemory and non-coherent caches, brings renewed attentionto the use of Software Transactional Memory (STM) asan alternative to lock-based synchronisation. However, STMrelies on the possibility of aborting conflicting transactions tomaintain data consistency, which impacts on the responsivenessand timing guarantees required by real-time systems. In thesesystems, contention delays must be (efficiently) limited so thatthe response times of tasks executing transactions are upperboundedand task sets can be feasibly scheduled. In this paperwe defend the role of the transaction contention managerto reduce the number of transaction retries and to help thereal-time scheduler assuring schedulability. For such purpose,the contention management policy should be aware of on-linescheduling information.
展开▼