ADAPTIVE MULTI-CRITERIA OPTIMIZATION OF RTCT VERIFICATION IN DECENTRALIZED OVERLAY NETWORKS

Abstract

Decentralized overlay networks – peer-to-peer systems, Internet-of-Things infrastructures, mesh and MANET environments, and the peer layer of Web3 ecosystems – operate without a single coordination center or trusted certification authority, which makes guaranteed verification of participant uniqueness impossible and opens the way to the Sybil attack and its derivatives: distributed-storage table poisoning, node eclipsing, artificial reputation inflation, and retention of an advantageous routing position. The mechanism of random time challenge tokens (RTCT) shifts protection from one-off admission control to participation-retention control; however, its practical suitability is determined by the choice of verification parameters – puzzle difficulty, challenge intensity, and response-window width. Baseline deterministic difficulty-adjustment schemes control a single parameter, react only to aggregate timing deviation, and are insensitive to the hardware heterogeneity of nodes. This article improves the method of adaptive tuning of participation-control parameters: the approach moves from adjusting difficulty alone to analytical multi-criteria optimization of the RTCT verification triple under the criteria of attack resistance and the computational and communication overhead of honest nodes. The problem is scalarized by a weighted sum whose weights parameterize the Pareto frontier. The optimum is characterized by the Karush–Kuhn–Tucker conditions under the assumptions of concavity of the security index and convexity of the overhead functions, which guarantees a unique global solution independent of the search algorithm. Analysis of the first-order conditions yields a meaningful tuning principle: the response window is tightened only as far as the honest-node false-positive budget permits, while intensity and difficulty are set by equalizing the marginal gain in resistance per unit of overhead. The linear scaling of the adversary’s resources with the number of fake identities is justified by queueing theory. A model illustrative example demonstrates the interior optimum and its comparative statics, consistent with the requirements of adaptivity. For non-stationary regimes, an optional realization of parameter selection as a Markov decision process solved by reinforcement learning is outlined, which is built on top of the analytical optimum and does not weaken the provability of the guarantees.