Self-Adaptation in Large-Scale Systems: A Study on Structured Overlays across Multiple Datacenters

With the recent focus on cloud computing a new type of system topology came up: clusters in geographically distributed datacenters that are connected by high-latency networks. Current structured overlay networks (SONs) are not well prepared for such environments with heterogeneous network performance and correlated node failures. We show how the beneficial features of SONs, namely self-management, scalability, adaptability, and fault tolerance can be exploited for multi-datacenter environments. We present self-adaptive replica placement policies and latency-optimized routing for SONs on multiple datacenters. Empirical results of our gossipbased ring maintenance protocol demonstrate its ability to cope with correlated node failures and network partitioning.