Online Allocation of Communication and Computation Resources for Real-time Multimedia Services

We observe that, in a network, the location of the node on which a service is computed is inextricably linked to the locations of the paths through which the service communicates. Hence service location can have a profound effect on quality of service (QoS), especially for communication-centric applications such as real-time multimedia. In this paper, we propose an online algorithm that uses pricing to consider server load, route congestion, and propagation delay jointly when locating servers and routes for real-time multimedia services in a network with fixed computing and communication capacities. The algorithm is online in the sense that it is able to sequentially allocate resources for services with long and unknown duration as demands arrive, without benefit of looking ahead to later demands. By formulating the problem as one of lowest cost subgraph packing, we prove that our algorithm is nevertheless C-competitive with the optimal algorithm that looks ahead, meaning that our performance is within a constant factor C of optimal, as measured by the total number of service demands satisfied, or total user utility. Using mixing services as an example, we show through experimental results that our algorithm can adapt to cross traffic and automatically route around congestion and failure of nodes and edges, can reduce latency by 40% or more, and can pack 20% more sessions, compared to conventional approaches.