Explicit Spatial Scattering for Load Balancing in Conservatively Synchronized Parallel Discrete Event Simulations

Abstract

Abstract-We re-examine the problem of load balancing in conservatively synchronized parallel, discrete-event simulations executed on high-performance computing clusters, focusing on simulations where computational and messaging load tend to be spatially clustered. Such domains are frequently characterized by the presence of geographic "hot-spots" -regions that generate significantly more simulation events than others. Examples of such domains include simulation of urban regions, transportation networks and networks where interaction between entities is often constrained by physical proximity.Noting that in conservatively synchronized parallel simulations, the speed of execution of the simulation is determined by the slowest ( i.e most heavily loaded) simulation process, we study different partitioning strategies in achieving equitable processor-load distribution in domains with spatially clustered load. In particular, we study the effectiveness of partitioning via spatial scattering to achieve optimal load balance. In this partitioning technique, nearby entities are explicitly assigned to different processors, thereby scattering the load across the cluster. This is motivated by two observations, namely, (i) since load is spatially clustered, spatial scattering should, intuitively, spread the load across the compute cluster, and (ii) in parallel simulations, equitable distribution of CPU load is a greater determinant of execution speed than message passing overhead. Through large-scale simulation experiments -both of abstracted and real simulation modelswe observe that scatter partitioning, even with its greatly increased messaging overhead, significantly outperforms more conventional spatial partitioning techniques that seek to reduce messaging overhead.Further, even if hot-spots change over the course of the simulation, if the underlying feature of spatial clustering is retained, load continues to be balanced with spatial scattering leading us to the observation that spatial scattering can often obviate the need for dynamic load balancing.

Parallelizing a discrete event simulation application using the Habanero-Java multicore library

“…In conservative algorithms [9,10,25], each LP processes its local events strictly in timestamp order. Each time an LP runs, it is allowed to process only safe events, which are local events that have timestamps smaller than any events the LP may receive in the future; otherwise, the LP must block itself, which may cause deadlocks in the system.…”

“…Similar to the workset implementation in the sequential version, we do not need redundant nodes in the task deques inside HJlib. Also, for each node which the active node is trying to spawn a new task for (line [18][19][20][21][22][23][24][25][26][27], if the node has one or more locks held by others at the same time, the new task does not need to be spawn for the node (line 26 does not executed in this case). One of the nodes that are still holding a lock of the node will spawn a new task for the node later.…”

On the distributed orchestration of stochastic discrete event simulations

“…Thulasidasan et al have focused on the load balancing optimization [15]. Thulasidasan et al have focused on the load balancing optimization [15].…”

Partitioning on Dynamic Behavior for Parallel Discrete Event Simulation

“…Thulasidasan et al [27] argue that dynamic load balancing presents significant implementation challenges due to object migration and explores the possibility of static partitioning for conservative simulation specifically for spatially clustered models with geographic hot-spots where most of the computation and messaging occurs (e.g. urban regions in transportation networks).…”