Optimizing capacity-heterogeneous unstructured P2P networks for random-walk traffic

Abstract

Existing algorithms for utilizing high-capacity nodes in heterogeneous P2P systems (e.g., [4], [13], [20]

On the tradeoff between resilience and degree overload in dynamic P2P graphs

“…The latter uses all links bidirectionally, benefitting from both incoming and outgoing edges, and never replaces those that fail. Recent work [39] showed initial evidence suggesting that passive networks were highly appealing as they exhibited good levels of disconnection resilience, in addition to simpler operation and lower overhead compared to active networks [22]. We therefore continue investigation in this direction and limit analysis to these types of P2P graphs.…”

“…Age-biased neighbor preference can be implemented in two general frameworks -active [5], [15], [23], [27], [34] and passive [14], [22]. The former relies on k out-neighbors for routing/resilience and uses edge rewiring upon detecting outneighbor departure.…”

“…Note that (22) is an exponential distribution if a = 1, heavy-tailed (NWU) if a < 1, and light-tailed (NBU) when a > 1. Weibull's age is distributed according to:…”

On superposition of heterogeneous edge processes in dynamic random graphs

“…Previous research has analyzed multiple avenues for understanding and improving such networks, which includes ensuring connectivity [11], balancing load [27], reducing graph diameter [17], improving resilience [14], [31], mitigating channel interference [1], understanding routing mobility [25] and flooding [26], and optimizing capacity [4], [9], [19]. While relying on a separate model for each study is acceptable in certain cases, the field of dynamic distributed systems has reached sufficient maturity that calls for a unifying foundation that can explain the limiting behavior of the aggregate edge process of the system and pave the way for rigorous analysis of diverse application-specific metrics.…”

“…k and θ are in (19), and l is the mean lifetime given in (13). The local rate γZ v (t) in Theorem 2 means that when user v remains alive in the system (i.e., Z v (·) = 1), the instantaneous rate of edge arrival to v is the constant γ; otherwise it is 0 (i.e., no users connect to v when it is offline).…”