Fully Sequential Procedures for Large-Scale Ranking-and-Selection Problems in Parallel Computing Environments

Abstract

Fully sequential ranking-and-selection (R&S) procedures to find the best from a finite set of simulated alternatives are often designed to be implemented on a single processor. However, parallel computing environments, such as multi-core personal computers and many-core servers, are becoming ubiquitous and easily accessible for ordinary users. In this paper, we propose two types of fully sequential procedures that can be used in parallel computing environments. We call them vector-filling procedures and asymptotic parallel selection procedures, respectively. Extensive numerical experiments show that the proposed procedures can take advantage of multiple parallel processors and solve large-scale R&S problems.

‘Some tactical problems in digital simulation’ for the next 10 years

“…This implies that the simulation results from multiple processors observed in the order that they are completed are not identically distributed (and, as it turns out, not independent either). Heidelberger (1988) and Glynn and Heidelberger (1991) only considered a single scenario with replications assigned to multiple processors, but the statistical problems get worse when there are multiple scenarios (Luo et al, 2015).…”

“…Efficiently use the output data in the order that it is returned, but derive procedures that are statistically valid anyway, as in Luo et al (2015).…”

A comparison of two parallel ranking and selection procedures

“…Thus it is natural to consider creating parallel versions of fully (or nearly-fully) sequential procedures. Indeed, recent work on parallel R&S procedures includes algorithms designed to sequentially eliminate systems from consideration via periodic screening on a parallel platform (Luo and Hong 2011;Luo et al 2013;Ni, Hunter, and Henderson 2013). However, when moving from a serial platform to a parallel one, new challenges arise that call into question the relative efficiency of nearly-fully sequential parallel procedures over naively parallelized two-stage procedures for some problem types.…”

“…A more detailed description and sample simulation code in both Matlab and C++ in for this problem can be found on SimOpt.org (Pasupathy and Henderson 2011). A number of prior studies including Pichitlamken, Nelson, and Hong (2006), Ni, Hunter, and Henderson (2013) and Luo et al (2013) have used different R&S approaches to solve a small version of the problem with R = B = 20 and k = 3249 systems.…”

Simulation optimization: A panel on the state of the art in research and practice

“…It is not just a case of "giving one replication to each core and gathering the results." Indeed, there are many algorithmic and statistical complexities involved; see Luo, Hong, Nelson, andWu (2013) andNi, Hunter, andHenderson (2013). I should also emphasize that I am thinking of SO problems where the individual simulation replications can be easily completed on a single core, as opposed to the setting in "Parallel and Distributed Simulation" where a single replication requires large numbers of cores to complete.…”