On Improving Transactional Memory: Optimistic Transactional Boosting, Remote Execution, and Hybrid Transactions

(ABSTRACT) Transactional memory (TM) has emerged as a promising synchronization abstraction for multi-core architectures. Unlike traditional lock based approaches, TM shifts the burden of synchronization from the programmer to an underlying framework using hardware (HTM) and/or software (STM) components. Although STM provides a generic solution for developing more complex concurrent applications , it suffers from scalability challenges due to the intensive speculation and meta-data handling. Moreover, In specific components such as transactional data structures, this intensive speculation results in more false conflicts than the optimized concurrent data structures. On the other hand, as computer architects and vendors chose the design of simple best-effort HTM components, HTM imposes limitations on the transactions and provides no guarantees for the transactions to eventually commit in hardware. These limitations raise the need to move into two (orthogonal) directions. The first direction is to design efficient hybrid TM frameworks which minimize the overhead of any HTM/STM interconnection. The second direction is to enhance the performance of the fall-back STM paths themselves, including enhancing specific components such as transactional data structures. In this dissertation, we propose two approaches to enhance the overall TM performance. First, we design an optimistic methodology for transactional boosting to specifically enhance the performance of the transactional data structures. Second, we enhance the performance of STM in general by dedicating hardware cores for executing specific parts of the transactions like commit and invalidation. In both approaches, we first show a pure software solution. Then we propose a hybrid TM solution which exploits the new HTM features of the currently released Intel's Haswell processor. Optimistic transactional boosting (OTB) is a methodology to design a transactional versions of the highly concurrent lazy data structures. An earlier (pessimistic) boosting proposal adds a layer of abstract locks on top of the optimized concurrent data structures. We propose an optimistic methodology of boosting which allows greater data structure-specific optimizations, easier integration with STM frameworks, and lower restrictions on the boosted operations than the original pessimistic boosting methodology. Based on our proposed OTB methodology, we implement transactional versions of both set and priority queue. We also extend the design of DEUCE, a Java STM framework, to support OTB integration. Using our extension, programmers can include both OTB data structure operations and traditional memory reads/writes in the same transaction. Results show that OTB performance is closer to the optimal lazy (non-transactional) data structures than the original boosting algorithm. Remote …