Numerous massively multiplayer online games (MMOGs) have come out in recent years, and these systems have very interesting performance requirements. These applications need to guarantee fast response times to a very dynamically varying number of clients, while simultaneously controlling and updating the simulated virtual environment. In order to satisfy these requirements, most providers rely on divide-and-conquer schemes and have a static resource pool big enough to cover peak demands; this leads, most of the time, to bad resource utilization.
Jordan et al. describe research and experiments for an architecture that would adapt dynamically to the workload. Essentially, the architecture consists of two layers of balancers: local balancers that shift loads inside data centers and global balancers that shift loads between data centers. Both layers use a bin-packing algorithm, which is described in the paper. Experimental evidence, obtained from simulations with recorded data from real MMOGs, indicates that the proposed architecture has better resource utilization than the static approach.
The paper is well presented, and the architecture is certainly relevant for MMOG implementers. For the most part, the referenced work is concerned with MMOGs. It would have been interesting to include a comparison to architectures from other application domains, such as cloud and grid architectures.