Towards Symmetric Multi-threaded Optimistic Simulation Kernels

Abstract

In this article we address the reshuffle of the design of optimistic simulation kernels in order to fit multi-core/multiprocessor machines. This is done by providing a reference optimistic simulation architecture based on the symmetric multithreaded paradigm, where each simulation kernel instance is allowed to run a dynamically changing set of worker threads that share the whole load of LPs hosted by that kernel, and that can run both application-level event handlers and kernellevel housekeeping tasks. With this organization, CPU-cores can be dynamically reassigned to the different kernels depending on fluctuations of the workload, so to maximize productivity in an orthogonal manner with respect to traditional load balancing schemes, typically employed in the context of single-threaded simulation kernels. In order to optimize efficiency and reduce wait-for-lock-release phases while synchronizing worker threads running in kernel mode, we borrow from Operating Systems' theory by readapting the top/bottom-halves paradigm to the design of optimistic simulation systems. We also present a real implementation of our multi-threaded architecture within the ROme OpTimistic Simulator (ROOT-Sim), namely an opensource C-based simulation platform implemented according to the PDES paradigm and the optimistic synchronization approach. Experimental results for an assessment of the validity of our proposal are presented as well.

Load-Sharing Policies in Parallel Simulation of Agent-Based Demographic Models

“…We consider the availability of C cores, and complying with the organization in [16], we assume K worker threads (K ≤ C) are available for event processing. To determine what LPs should be bound to the available worker threads, we follow these steps:…”

“…Our goal is to give the highest degree of freedom to the programmer, and to ensure an efficient execution of the simulation. We target symmetric-multithread PDES environments for shared-memory multicore systems [16]. LPs are allowed to interact in a twofold way: (i) explicitly, namely via traditional message passing, or (ii) implicitly, i.e.…”

“…As discussed in [16], this is a fundamental aspect to offer competitive performance. The binding temporarily assigns computing resources (i.e., worker threads stick to certain cores) to groups of LPs.…”

“…In the context of PDES, several works have studied the problem of finding the best binding between LPs and worker threads-see, e.g., [16,25,26,21,27]. Nevertheless, none of these works has ever used information related to the interaction between LPs to explicitly reduce the (possible) negative effects of optimistic simulation runs.…”

Exploring many-core architecture design space for parallel discrete event simulation

“…Vitali et al [32,31] developed a different multi-threaded PDES simulator on multi-core platforms. A load-sharing scheme was implemented, allowing each simulation kernel instance to be executed by multiple threads.…”