Transactional Memory Based Speculative Parallel Execution of Discrete Event Applications

Mauro Ianni

Abstract:
Speculative parallel processing is a well known means to deliver high performance and scalability when executing discrete event simulation models. Nevertheless, it requires the runtime support to restore the application’s state to some past (consistent) image. Traditionally, the recoverability support has been realized via proper software layers. However, although a lot of optimizations have been provided in literature for making software-based recoverability highly efficient, its relative overhead may still represent an impairment to performance in case of (very) fine grain applications. This work presents an innovative runtime support for speculative parallel processing of discrete event simulation models on multi-core architectures, which exploits Hardware-Transactional-Memory (HTM) facilities, nowadays offered by offthe-shelf processors, for the purpose of state recoverability. In this thesis, the speculative updates on the state of the simulation model are executed as concurrent HTM-based transactions that are also in charge of detecting whether the update is consistent with the advancement of logical-time along model execution. This is achieved by including in the HTM-based transactional code-block both the activation of the application layer in charge of processing the simulation event, and the execution of housekeeping tasks aimed at determining the safety (in terms of causal consistency) of the executed transaction. This proposal is fully transparent to the application code. Hence, this HTM-based run-time support can host conventionally developed discrete event models relying on the concept of event-handlers to be dispatched by an underlying simulation engine. Experimental data show that this proposal provides 75% to 92% of the ideal speedup on an Intel Haswell based platform (equipped with 4 physical cores and HTM support) for discrete event models with event granularity ranging between 2 and 12 microseconds. The data also show that these same models cannot be executed efficiently on top of a last generation parallel discrete event simulation platform employing software-based recoverability

BibTeX Entry: