Online client assignment in dynamic real-time distributed interactive applications

Abstract

Qulaity of user experience in Distributed Interactive Applications (DIAs) highly depends on the network latencies during the system execution. In DIAs, each user is assigned to a server and communication with any other client is performed throught its assigned server. Hence, latency measured between two clients, called interaction time, consists of two components. One is the latency between the client and its assigned server, and the other is the inter-server latency, that is the latency between servers that the clients are assigned. In this paper, we investigate a real-time client to server assignment scheme in a DIA where the objective is to minimize the interaction time among clients. The client assignment problem is known to be NP-complete and heuristics play an important role in finding near optimal solutions. We propose two distributed heuristic algorithms to the online client assignment problem in a dynamic DIA system. We utilized real-time Internet latency data on PlanetLab platform and performed extensive 3 experiments using geographically distributed PlanetLab nodes where nodes can arbitrarily join/leave the system. The experimental results demonstrate that our proposed algorthims can reduce the maximum interaction time among clients up to 45% compared to an exiting baseline technique.