This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Firefly-inspired Heartbeat Synchronization in Overlay Networks Ozalp Babaoglu Univ. Bologna, Italy firstname.lastname@example.org Toni Binci Univ. Bologna, Italy email@example.com Mark Jelasity HAS & Univ. Szeged, Hungary firstname.lastname@example.org Alberto Montresor Univ. Trento, Italy email@example.com Abstract Heartbeat synchronization strives to have nodes in a distributed system generate periodic, local heartbeat events approximately at the same time. Many useful dis- tributed protocols rely on the existence of such heart- beats for driving their cycle-based execution. Yet, solv- ing the problem in environments where nodes are unre- liable and messages are subject to delays and failures is non-trivial. We present a heartbeat synchronization protocol for overlay networks inspired by mathemati- cal models of flash synchronization in certain species of fireflies. In our protocol, nodes send flash messages to their neighbors when a local heartbeat triggers. They adjust the phase of their next heartbeat based on in- coming flash messages using an algorithm inspired by mathematical models of firefly synchronization. We re- port simulation results of the protocol in various real- istic failure scenarios typical in overlay networks and show that synchronization emerges even when messages can have significant delay subject to large jitter. 1. Introduction In cycle- or round-based distributed protocols (such as gossip protocols), it is often necessary that all nodes agree on when a new cycle starts. In other words, the lo- cal perceptions at nodes as to when cycles begin and end Authors are listed in alphabetical order. Partial support for this work was provided by the European Union within the 6th Framework Programme under contracts 001907 (DELIS) and 27748 (BIONETS). need to be synchronized so that we can talk about cy- cles of the system as a whole. For example, if the pro- tocol requires periodic restarts (that is, all nodes need to be re-initialized), it is important that this event be syn- chronized [7,9]. Heartbeat synchronization strives to have nodes in a distributed system generate periodic, local heartbeat events approximately at the same time. It differs from classical clock synchronization in that nodes are not in- terested in counting cycles and agreeing on the ID of the current cycle. Furthermore, there is no requirement regarding the length of a cycle with respect to real time as long as the length is bounded and all nodes agree on it eventually. What we are interested in guaranteeing is that all nodes start and end their cycles at the same time, with an error that is at least one, but preferably more, or- ders of magnitude smaller than the chosen cycle length....
View Full Document
- Spring '08