Migration-Based Synchronization

Abstract

A fundamental challenge in multi- and many-core systems is the correct execution of concurrent access to shared data. A common drawback from existing synchronization mechanisms is the loss of data locality as the shared data is transferred between the accessing cores. In real-time systems, this is especially important as knowledge about data access times is crucial to establish bounds on execution times and guarantee the meeting of deadlines.We propose in this paper a refinement of our previously sketched approach of Migration-Based Synchronization (MBS) as well as its first practical implementation. The core concept of MBS is the replacement of data migration with control-flow migration to achieve synchronized memory accesses with guaranteed data locality. This leads to both shorter and more predictable execution times for critical sections. As MBS can be used as a substitute for classical locks, it can be employed in legacy applications without code alterations.We further examine how the gained data locality improves the results of worst-case timing analyses and results in tighter bounds on execution and response time. We reason about the similarity of MBS to existing synchronization approaches and how it enables us to reuse existing analysis techniques.Finally, we evaluate our prototype implementation, showing that MBS can exploit data locality with similar overheads as traditional locking mechanisms.