The Challenges of Merging Two Similar Structured Overlays: A Tale of Two Networks

Structured overlay networks is an important and interesting primitive that can be used by diverse peer-to-peer applications. Multiple overlays can result either because of network partitioning or (more likely) because different groups of peers build such overlays separately before coming in contact with each other and wishing to coalesce the overlays together. This paper is a first look into how multiple such overlays (all using the same protocols) can be merged – which is critical for usability and adoption of such an internet-scale distributed system. We elaborate how two networks using the same protocols can be merged, looking specifically into two different overlay design principles: (i) maintaining the ring invariant and (ii) structural replications, either of which are used in various overlay networks to guarantee functional correctness in a highly dynamic (membership changes) environment. Particularly, we show that ring based networks can not operate until the merger operation completes. In contrast, from the perspective of individual peers in structurally replicated overlays there is no disruption of service, and they can continue to discover and access resources that they could originally do before the beginning of the merger process, even though resources from the other network become visible only gradually with the progress of the merger process.