Enabling fine-grained load balancing for virtual worlds with distributed simulation engines
Arthur Valadares, Cristina Videira Lopes, Huaiyu Liu · 2014
Virtual worlds are general-purpose real-time simulation of three-dimensional environments, and serve for several purposes, such as physics simulation, collaboration, and entertainment. Due to the real-time nature of these simulations, scaling the number of in-world entities and interacting users is challenging. In this paper we present a novel approach to scalable virtual worlds, combining two dimensions of workload partitioning: space and operations. We present this new design as the Distributed Scene Graph with microcells (DSG-M), and evaluate our approach in a distributed physics intensive evaluation aimed at testing two hypothesis: (1) the space partitioning approach improves scalability by balancing the load of an overwhelmed physics dedicated simulator; and (2) simulation precision can be maintained by assigning read-only spaces near the partition borders. Results show evidence to confirm both hypotheses, and of successfully scaling the simulation of an overwhelming workload.