Effects of Replica Placement Algorithms on Performance of structured Overlay Networks

In DHT-based P2P systems, replication-based content distribution and load balancing strategies consists of such decisions as which files should be replicated, how many replicas should be created and where to replicate them in order increase the system performance in the presence of non-uniform data and access distribution. There are many works on replica placement policies; however, the impact of system workload on different replica placement strategies is not well studied. We investigate this problem under the context of content addressable overlay networks. We compare a trace based replica placement algorithm with two of its variations, namely random placement and priority based placement under different workloads. Our experimental results show that the effect of replica placement policy is highly affected by the workload of the system, which indicates that an adaptive replica placement strategy is desirable for content distribution in an overlay network.