Transactional Memory (TM), an attractive solution to support concurrent accesses to main-memory storage, is already being deployed by some of the major CPU and compiler manufacturers. To address scalability and dependability challenges, researchers are now combining replication, TM and certification-based protocols. To maintain consistency and ensure common transaction serialisation order, these protocols rely in a total-order broadcast primitive, usually provided by some Group Communication System(GCS). The total-order broadcast service can be implemented by different algorithms, which hold different properties. In this paper we present a detailed analysis of the impact of some algorithms implementing total-order broadcast in different TM workloads, opening up future work to improve performance of replicated TMs.
|Title of host publication||Lecture Notes in Computer Science|
|Publication status||Published - 1 Jan 2013|
|Event||Multicore Software Engineering, Performance, and Tools - |
Duration: 1 Jan 2013 → …
|Conference||Multicore Software Engineering, Performance, and Tools|
|Period||1/01/13 → …|