Reshuffling PDES Platforms for Multi/Many-core Machines: a Perspective with focus on Load Sharing

Simulation is an attractive and well-consolidated methodology to study real-world phenomena. In particular, parallel discrete event simulation (PDES) is a paradigm that has been extensively used (because of its modular and simple way of specifying simulation models) and has been proven as very effective in a wide set of fields, including physics, biology, and business-oriented processes (such as financial prediction or optimized system-configuration selection). A core aspect of PDES platforms is that they allow the exploitation of multiple computing units to give rise to a parallel execution of the simulation model, which can provide significant reduction of the time requested for delivering simulation outputs to end users or applications. Specifically, PDES has been shown to provide (large) speedups when compared to classical simulation paradigms where the execution of the simulation model takes place in a sequential fashion. This enables the wide usage of simulation in contexts where timeliness in the production of the simulation results plays a fundamental role, such as when employing simulation technology in time-critical decision processes. The PDES paradigm dates back to the 80s and was originally thought as of a means for exploiting computing systems formed by clusters of machines relying on single-core central processing unit (CPU) technology. On the other hand, more recent technological trends lead to the proliferation and large diffusion of multicore hardware, where multiple computing units are hosted on the same machine and share several hardware resources, such as main memory. This unavoidably leads to the need for rethinking the organization of PDES platforms to make them perfectly suited for exploiting the computing power offered by modern multi/many-core machines. In this chapter, we discuss some key aspects related to the reorganization process of these platforms and present in detail a recent literature approach exactly tackling this issue. The presentation is also targeted at showing how the approach, which is based on the symmetric multithreading software programming paradigm, can be suited for a change in the perspective on how to exploit computing resources for PDES applications in a balanced and effective manner. This is achieved via an innovative load-sharing paradigm suited for PDES systems run on top of multicore machines. The chapter also considers a case study in the context of optimistically synchronized PDES platforms, which are based on speculative processing schemes and achieve causal consistency of the parallel run by means of rollback/recovery techniques. The case study reports the main outcomes of an experimental assessment of the suitability of the revised organization of PDES systems and of the load-sharing technique. The remainder of this chapter is organized as follows. We initially provide background information in relation to the PDES paradigm, including some examples of PDES-compliant models, which may help the reader to fully understand the paradigm potential. Then we enter details related to the multicore technological trend, also discussing the motivations for its adoption. Afterwards, we outline the main challenges and opportunities related to the organization of modern PDES platforms to be run on top of multicore machines. Successively, we provide an overview of the aforementioned recent proposal particularly aimed at achieving balanced exploitation of the available computing resources when running PDES models on multicore systems. Finally, we present the case study on the design and implementation of an instance of PDES platform tailored for multicore machines, based on the optimistic synchronization paradigm, and report some results related to its run-time behavior.