An analysis of queuing network simulation using GPU-based hardware acceleration

Abstract

Queuing networks are used widely in computer simulation studies. Examples of queuing networks can be found in areas such as the supply chains, manufacturing work flow, and internet routing. If the networks are fairly small in size and complexity, it is possible to create discrete event simulations of the networks without incurring significant delays in analyzing the system. However, as the networks grow in size, such analysis can be time consuming, and thus require more expensive parallel processing computers or clusters. We have constructed a set of tools that allow the analyst to simulate queuing networks in parallel, using the fairly inexpensive and commonly available graphics processing units (GPUs) found in most recent computing platforms. We present an analysis of a GPU-based algorithm, describing benefits and issues with the GPU approach. The algorithm clusters events, achieving speedup at the expense of an approximation error which grows as the cluster size increases. We were able to achieve 10-x speedup using our approach with a small error in a specific implementation of a synthetic closed queuing network simulation. This error can be mitigated, based on error analysis trends, obtaining reasonably accurate output statistics. The experimental results of the mobile ad hoc network simulation show that errors occur only in the time-dependent output statistics.

GPU accelerated three-stage execution model for event-parallel simulation

“…Many works focus on the issues in adopting the GPU platform in the applications [1,7,16,15,14,13]. The branch divergency of threads, memory hierarchy on GPU, host-device data transfer latency and global memory access pattern are mostly discussed in these works.…”

“…In the work of Park and Fishwick [14,13], the authors propose more generic data structures to process parallel DES. The future event list (FEL) is decomposed into sub-FELs then assigned to GPU threads for concurrent processing.…”

“…To execute DES on GPU, in Park and Fishwick's work [14], at the end of each simulation loop, the future event list (FEL) on GPU will be copied back to the host side memory to either double the array size of the FEL or cut it in half, based on the number of elements in the FEL. By adjusting at the host side, the size of the FEL can be dynamic.…”

A GPU-Based Simulation Kernel within Heterogeneous Collaborative Computation on Large-Scale Artificial Society

“…For example, Perumalla et al, experiment on several typical agent-based simulation application in GPU, and obtain large speedup compared to that in CPU [6], and he also realize the simulation of diffusion system in GPU by the event driven and time step discrete event scheduling algorithm for the first time [7]. Hyungwook Park et al, present a framework of discrete event simulation application based on GPU [8], which divide the input event the queue into several sub-queue to enhance the degree of parallelism of event executing in GPU, it achieves good execution performance in queue simulation system [9]. Wenjie Tang, Yiping Yao et al, realizes a general GPU-based discrete event simulation kernel, and proposes an expansion-aided synchronous conservative time management algorithm, and a memory management algorithm, which solve the problem of memory access conflict in GPU parallel process, and raise the parallelism of simulation kernel by adding more concurrent events [10], [11].…”

The Speedup of Discrete Event Simulations by Utilizing CPU Caching

“…As an instance of attempts, recently H. Park and P.A. Fishwick proposed a graphics processing units (GPU) based algorithm which shows 10-fold speedup [11]. However, until now the caching mechanism of processor has not been considered in existing event scheduling algorithms.…”