Large-Scale Service Overlay Networking with Distance-Based Clustering

Abstract

Abstract. The problem of service routing (or dynamic service composition) has recently emerged as a consequence of the distributed composable services model residing in middleware layer(s). However, existing solutions are mostly suitable for small-or medium-scale service overlay networks, as service routing is performed over flat overlay topologies such as a mesh. Due to their increasing routing information maintenance costs, these flat (single-level) topology solutions cannot cope with largescale service overlay networking. For better scalability, in this paper, we provide a hierarchical service routing framework, which comprises three parts. In the first part, we organize the overlay network nodes into clusters based on their Internet distances. We then construct a hierarchically fully connected (HFC) topology based on the clustering result. In such a topology, nodes within a cluster are considered fully connected, and the clusters themselves are also fully connected by their border nodes. In the second part, a hierarchical state information distribution protocol will be provided so that each node in the system maintains full state of the nodes in its own cluster and aggregate state of other clusters in the system. In the third part, we present how service paths can be computed hierarchically in a divide-and-conquer fashion. Through simulation tests, we demonstrate that while achieving much better scalability, our framework provides also as good and efficient service paths as single-level mesh solutions.

On Exploring Performance Optimizations in Web Service Composition

“…A remedial step for increasing scalability is to introduce hierarchies into the network, so that topology abstraction and state information become possible to significantly reduce the state maintenance overhead. A hierarchical solution was developed in [6]. Alternatively, scalable service routing can take a distributed approach by having the network nodes maintaining state information of a limited neighborhood.…”

“…Service unicast routing has been reasonably addressed in the literature [6][7][8][9]. Some of the existing work, e.g., [8,9], adopt a global planning approach which, concerning its limited scalability, is not suitable for the current Web.…”

“…Some of the existing work, e.g., [8,9], adopt a global planning approach which, concerning its limited scalability, is not suitable for the current Web. Scalable routing falls into two approaches: hierarchical [6] and distributed [7], each with its own advantages and disadvantages. The routing approach to be adopted in this paper follows the latter category, because distributed routing based on on-line probing involves with more updated (thus more accurate) routing state.…”

Hybrid multicasting in large-scale service networks

“…Service unicast has been reasonably addressed in the literature [1,2,3]. Some of the existing work, e.g., [3], adopts a global planning approach which, concerning its limited scalability, is not This work was supported by NSF grants CCR-9988199 and EIA 99-72884 EQ, and NASA grant NAG2-1406.…”

“…suitable for large networks (e.g., the current Web). Scalable routing falls into two approaches: hierarchical [1] and distributed [2]. The routing approach in this paper falls into the latter category.…”