On Composability, Efficient Design and Memory Reclamation of Lock-free Data Structures

The transition to multicore processors has brought synchronization, a fundamental challenge in computer science, into focus. In looking for solutions to the problem, interest has developed in the lock-free approach, which has been proven to achieve several advantages over the traditional mutual exclusion approach. This thesis studies challenges in interprocess synchronization in shared memory multicore systems using the lock-free approach. Our contributions focus on efficient design and implementation, composition, and dynamic memory reclamation of lock-free data structures, a key component in lock-free solutions to synchronization problems. First, we show that lock-free synchronization offers several advantages. Lock-free implementations of data structures can achieve decent throughput performance while managing to provide competitive fairness among the sharing participants in accessing the shared data. We also show that although lockfreedom does not guarantee starvation-freedom, it is composable in terms of the progress guarantee. Multiple lock-free data objects can concurrently use another lock-free object without compromising their lock-free progress guarantees because operations they invoke at that object get starved. Having shown that lock-free synchronization possesses several advantages, we then propose lock-free implementations of data structures, as they play a vital role in solving synchronization problems. We present a lock-free hash table based on cuckoo hashing scheme and a lock-free skip-list with extended functionality. Cuckoo hashing uses two hash tables to offer two positions for any key, so hashing conflicts are solved efficiently and simply by placing conflicted keys in different positions. We develop a lock-free implementation by addressing challenges in manipulating elements in their two possible positions. The evaluation results show that our lock-free cuckoo hash table outperforms other