Load sharing for optimistic parallel simulations on multi core machines

Abstract

Parallel Discrete Event Simulation (PDES) is based on the partitioning of the simulation model into distinct Logical Processes (LPs), each one modeling a portion of the entire system, which are allowed to execute simulation events concurrently. This allows exploiting parallel computing architectures to speedup model execution, and to make very large models tractable. In this article we cope with the optimistic approach to PDES, where LPs are allowed to concurrently process their events in a speculative fashion, and rollback/recovery techniques are used to guarantee state consistency in case of causality violations along the speculative execution path. Particularly, we present an innovative load sharing approach targeted at optimizing resource usage for fruitful simulation work when running an optimistic PDES environment on top of multi-processor/multi-core machines. Beyond providing the load sharing model, we also define a load sharing oriented architectural scheme, based on a symmetric multi-threaded organization of the simulation platform. Finally, we present a real implementation of the load sharing architecture within the open source ROme OpTimistic Simulator (ROOT-Sim) package. Experimental data for an assessment of both viability and effectiveness of our proposal are presented as well.

NUMA Time Warp

“…We target optimistic PDES systems based on the multithread paradigm (rather than multi-process ones), which have been recently shown to be highly suited for shared memory platforms thanks to the possibility of optimizing aspects such as data exchange and balanced usage of the computing power [36,37] in application transparent manner. Also, we consider a convenient execution model where a simulation object, which we denote as SOBJi, is temporarily bound to a specific worker thread W Tj along a given wall-clock-time window.…”

“…for load-sharing purposes [36]) and when the worker threads are switched across CPU-cores operating in different NUMA nodes;…”

“…The latter approach is actuated according to an non-intrusive daemonbased solution, where a thread (similar in spirit to the Linux kswapd) is in charge of periodically checking whether a migration from a source to a destination NUMA node is requested for the memory pages of any simulation object, and in the positive case executes the migration non-intrusively. Hence, our proposal guarantees NUMA-efficient memory access even in contexts where a specific simulation object is migrated from one worker thread to another one (possibly operating on a CPU-core close to a different NUMA node), such as for the case of load-sharing with dynamic migration of the objects within a multi-thread Time Warp platform [36] or when the number of worker threads is dynamically changed with consequent re-distribution of the simulation objects across the still active worker threads [38]. Further, our proposal entails optimized NUMA management also in relation to the memory buffers used for exchanging events across the concurrent simulation objects.…”

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

“…This can improve the usage of computing resources while carrying out speculative processing of DES models, by reducing negative effects of speculation, such as the rollback frequency. This is the objective of classical load balanging/sharing approaches proposed in literature (see, e.g., [19][20][21]). However, these proposal consider only explicit interactions supported via the classical event cross-scheduling approach.…”

“…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.…”