More scalability at lower costs – Server Architecture for Massive Multiplayer 3D Virtual Spaces powered by GPGPU

Abstract—3D massive multiplayer virtual spaces are getting more and more popular, not only as computer games but as complex simulation and interaction environments, heading to become the next paradigm of multi-user interface. Still their universal adoption is hindered by some serious practical issues, mainly revolving around development costs and scalability limitations. The authors consider that the main cause for these limitations resides in the particularities of server-side software architectures- traditionally designed as clusters of single processor machines. The paper gives a brief overview of current solutions and their limitations and proposes two innovative architectural concepts which have a big potential for creating cheaper and more scalable solutions. We describe a region based decomposition of the virtual space together with supporting middlewares of messaging, distributed control and persistence, which allow an efficient and flexible work effort distribution on server side. The solution allows for both horizontal and vertical scalability. The vertical scalability is then mapped in an innovative manner on the last generation of SIMD-like multi-processor graphics cards. The huge processing power of these cards, with the right architecture, can take over the bulk of the server-side effort. Our prototype tests indicated that the solution is feasible and may represent an important turnaround in the development of more scalable and much cheaper massive multi-player server architectures for various types of virtual spaces. Keywords—3D, virtual space, massive multiplayer, serve