Perigee: Efficient Peer-to-Peer Network Design for Blockchains

Abstract

A key performance metric in blockchains is the latency between when a transaction is broadcast and when it is confirmed (the socalled, confirmation latency). While improvements in consensus techniques can lead to lower confirmation latency, a fundamental lower bound on confirmation latency is the propagation latency of messages through the underlying peer-to-peer (p2p) network (in Bitcoin, the propagation latency is several tens of seconds). The de facto p2p protocol used by Bitcoin and other blockchains is based on random connectivity: each node connects to a random subset of nodes. The induced p2p network topology can be highly suboptimal since it neglects geographical distance, differences in bandwidth, hash-power and computational abilities across peers. We present Perigee, a decentralized algorithm that automatically learns an efficient p2p topology tuned to the aforementioned network heterogeneities, purely based on peers' interactions with their neighbors. Motivated by the literature on the multi-armed bandit problem, Perigee optimally balances the tradeoff between retaining connections to known well-connected neighbors, and exploring new connections to previously-unseen neighbors. Experimental evaluations show that Perigee reduces the latency to broadcast by 33%. Lastly Perigee is simple, computationally lightweight, adversary-resistant, and compatible with the selfish interests of peers, making it an attractive p2p protocol for blockchains.

Strategic Latency Reduction in Blockchain Peer-to-Peer Networks

“…Peri: A recently-proposed protocol called Perigee [29] leverages agents' ability in blockchain P2P networks to control their peering in order to achieve systemic, i.e., network-wide, latency improvements. In the Perigee protocol, every node favors peers that relay blocks quickly and rejects peers that fail to do so.…”

Less is More: Fairness in Wide-Area Proof-of-Work Blockchain Networks

“…Kadcast [35] proposes a structured p2p overlay with UDP and forward error correction to achieve a more efficient broadcast compared to Bitcoin's random topology with TCP links. Perigee [26] presents an algorithm to explore and exploit neighbors to search for the best set of neighbors to connect to, for optimizing broadcasting delay. Other innovations, e.g., Bitcoin's compact block relay [17] also help to reduce broadcast time by reducing the size of each block.…”

“…For a fixed hashrate, a miner may also be able to increase rewards by being well-connected in the p2p network and/or being favorably located in the network to send and receive blocks faster than other competing miners on average [19,21,41]. Recent works have explored various ways including topology redesigns [26,33,35], fast relay networks [10,11,23], and coding or compressing blocks [16,32] to reduce the latency of block propagation in the network. The conventional wisdom here is that a faster network that is able to disseminate blocks quickly is generally more desirable that a slower network, as it allows miners to hear about newly mined blocks promptly thus reducing the likelihood of mining new blocks over older blocks (a process called as forking §2).…”

Impact of delay classes on the data structure in IOTA

“…Messages are subject to several delay vectors before they can enter the tip pool of a node. First, the broadcast in the network does not happen instantly, but rather is dominated by a natural network delay that depends on several factors, such as geographical distance and distance in the P2P virtual network [10]. Second, messages also need to be processed locally, are members in a waiting pool before being processed, or depend on messages that have not yet arrived.…”