Vivaldi

Abstract

Large-scale Internet applications can benefit from an ability to predict round-trip times to other hosts without having to contact them first. Explicit measurements are often unattractive because the cost of measurement can outweigh the benefits of exploiting proximity information. Vivaldi is a simple, light-weight algorithm that assigns synthetic coordinates to hosts such that the distance between the coordinates of two hosts accurately predicts the communication latency between the hosts. Vivaldi is fully distributed, requiring no fixed network infrastructure and no distinguished hosts. It is also efficient: a new host can compute good coordinates for itself after collecting latency information from only a few other hosts. Because it requires little com-munication, Vivaldi can piggy-back on the communication patterns of the application using it and scale to a large number of hosts. An evaluation of Vivaldi using a simulated network whose latencies are based on measurements among 1740 Internet hosts shows that a 2-dimensional Euclidean model with height vectors embeds these hosts with low error (the median relative error in round-trip time prediction is 11 percent).

P2P matchmaking solution for online games

“…Htrae excels at combining both of those strategies into one. It works similarly to a network coordinate system called Vivaldi [30], that is, it assigns to each peer coordinates in a virtual spherical space (in Vivaldi the space was non-spherical). The clue of the system is that the initial positions in the virtual space are dictated by geolocation data but they are later adjusted based on latency measurements.…”

CHESS: An application-aware space for enhanced scalable services in overlay networks

“…Static approaches [2], [21] use metrics that change rarely over time as the number of hops, the domain name and the geographical location. Dynamic approaches [1], [21], [4] are based on the measurement of variable network metrics. They mainly focus on the delay and consider it as a measure of closeness of peers; the appropriate peer to contact is often taken as the closest one in the delay space.…”

“…The same algorithm has been applied in PIC [26], after investigating issues related to security. In [1], Vivaldi simulates the overlay network by a network of physical springs. It proposes to determine peers' coordinates in a distributed way without the need to dispose of a fixed set of landmarks.…”

“…Different functions are introduced in the literature to characterize the proximity of peers, but most of them [1], [2], [3], [4] are based on simple metrics such as the delay, the number of hops or the geographical location. We believe that these metrics are not enough to characterize the proximity given the heterogeneity of the Internet in terms of path characteristics and access link speed, and the diversity of application requirements.…”

“…To this end, we introduce the CHESS space where the proximity is characterized according 1 The notion of landmark in our context of bandwidth estimation is different from that used for delay estimation [3], [4], [8], [9]. Delay landmarks can be seen as reference points for inferring peer's network position.…”