Stabilizing Distributed R-Trees for Peer-to-Peer Content Routing

Abstract

Publish/subscribe systems provide useful platforms for delivering data (events) from publishers to subscribers in a decoupled fashion. Developing efficient publish/subscribe schemes in dynamic distributed systems is still an open problem for complex subscriptions (spanning multidimensional intervals). We propose a distributed R-tree (DR-tree) structure that uses R-tree-based spatial filters to construct a peer-to-peer overlay optimized for scalable and efficient selective dissemination of information. We adapt well-known variants of R-trees to organize publishers and subscribers in balanced peer-to-peer networks that support content-based filtering in publish/subscribe systems. DR-tree overlays guarantee subscription and publication times logarithmic in the size of the network while keeping space requirements low (comparable to distributed hash tables). The maintenance of the overlay is local and the structure is balanced with height logarithmic in the number of nodes. DR-tree overlays disseminate messages with no false negatives and very few false positives in the embedded publish/subscribe system. In addition, we propose self-stabilizing algorithms that guarantee consistency despite failures and changes in the peer population.

VCube-PS: A causal broadcast topic-based publish/subscribe system

“…Moreover, many of these multicast trees include unrelated intermediate hops and nodes that are not subscribers which have to forward the message presenting, thus, the problem of false positives and the need of message filtering (e.g. DR-Tree [22] and Scribe [3]). Finally, the maintenance cost is usually high, specially in presence of churn.…”

Expressive Content-Based Routing in Software-Defined Networks

“…The first step in the selective indexing approach is to select subspaces in Ω populated with subscriptions while identifying the empty spaces to be neglected. To identify meaningful subspaces, we benefit from the widely used mechanism of similarity-based subscription clustering [30], [10]. Once subscriptions are clustered into groups, we generate polyspace rectangles which serve as the closest enclosing approximation of each of these clusters.…”

A Publish/Subscribe System Using Causal Broadcast over Dynamically Built Spanning Trees

“…Similarly to VCube-PS, many Pub/Sub systems use treebased overlays (e.g., Scribe [3], Bayeux [4], Marshmallow [12], DR-Tree [8], DYNATOPS [6]). The advantage of using trees is the logarithmic guaranties on publication delivery time and the number of messages employed.…”

“…Moreover, many of these multicast trees include unrelated intermediate hops and nodes that are not subscribers which have to forward the message, presenting thus the problem of false positives and the need of message filtering (e.g. DR-Tree [8], Scribe [3]). Finally, their maintenance cost is usually high, specially in presence of churn.…”

“…There basically exist two models of Pub/Sub systems: topicbased [3], [4], [5], [6] and content-based [7], [8]. In the first one, subscribers share a common knowledge on a set of available topics and every published message is labeled with one of these topics.…”