IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 1, JANUARY 2004 1Guest EditorialRecent Advances in Service Overlay NetworksI. SERVICEOVERLAYNETWORKSTHE best effort Internet was designed when host connec-tivity was the primary concern. To achieve this purpose,the Internet followed a simple and robust design philosophy andhas become a phenomenal success in the last decade. Today’sInternet users are very different from those of a decade ago andtheir expectations with regard to the Internet have changed con-siderably. Users are now more concerned about timely deliveryof desired services than host connectivity. The role of the In-ternet has also evolved; the Internet has become a commercialinfrastructure for service delivery. However, due to the Internetprotocol (IP)’s addressing scheme, routing paradigm, and otherhistorical reasons, it is currently not well suited for the pur-pose of service delivery. In response to these challenges, variousforms of overlay networks have been proposed and some de-ployed over the Internet. For example, commercial content de-livery service providers have used proprietary server networksover the existing Internet to bring content closer to users. In themeantime, grassroots users have actively contributed their com-puters to form various kinds of peer-to-peer networks for fileswapping and content sharing.The purpose of this issue is to disseminate state-of-the-artresearch results that address service delivery over the Internetwith overlay networks.II. OVERVIEW OF THEISSUEThe 15 papers in this issue fall into the following areas: net-work construction, routing and discovery, measurement, overlaymulticasting, performance evaluations and design tradeoffs, se-curity and resilience.A. Network ConstructionThe first three papers in this special issue investigate archi-tectural issues in overlay networks. Internet flash crowds, alsoknown as “hot spots,” are a phenomenon that result from asudden, unpredicted increase in an online object’s popularity.When content reaches its apex in popularity, it becomes un-available to the majority of users who seek it. One approachto address this problem is to have clients form a peer-to-peer(P2P) overlay network that allows those clients who have re-ceived copies of popular content to forward it to those clientswho also desire but have not yet received it. The first paper inthis issue, “A Lightweight, Robust P2P System to Handle FlashCrowds,” describes one implementation of such an approach.This paper proposes to use randomized overlay constructionDigital Object Identifier 10.1109/JSAC.2003.818774and randomized, scoped searches to efficiently locate and de-liver objects that are under heavy demand to all users who de-sire them. The authors show via a mixture of theoretical results,simulation, and experimentation that by relying on randomness,their proposed P2P implementation can achieve low latency de-livery using modest traffic levels, even when membership to theoverlay changes dynamically with time and when there existmembers that limit their participation in the system.In the second paper, “A Construction of Locality-AwareOverlay Network: mOverlay and its Performance,” the au-thors propose to construct an overlay network by exploitingthe locality in the underlying network using group concept.This approach is based on the observation that a completedecoupling of overlay topology and underlying data networktopology may not be completely desirable. For instance, if theP2P overlay is constructed randomly, hosts that are nearbyin the overlay network could, in fact, be far away from theunderlying data network, which translates into inefficiency innetwork resource use and degradation in user-perceived per-formance. The authors propose a new protocol to construct anunstructured overlay network to achieve local characteristics.The third paper, “QRON: QoS-Aware Routing in OverlayNetworks,” envisions an overlay service network constructedby deploying so-called “overlay brokers” (OBs) in eachautonomous systems. By having these OBs cooperate witheach other, an overlay service network can perform variousnetwork services to support other applications, such as resourceallocation and routing. The authors focus on quality-of-service(QoS)-aware routing protocols within such overlay networks;they adopt a hierarchical approach in their design of routingalgorithms to achieve scalability. Through simulations, theauthors show that their routing algorithms can effectively findand provide QoS-satisfied overlay paths and can balance theoverlay traffic burden among OBs and overlay links.B. Routing and DiscoveryThe next two papers focus on routing and discovery forservice overlay networks. The fist paper, “Tapestry: A ResilientGlobal-Scale Overlay for Service Deployment,” presents a P2Poverlay routing infrastructure offering location-independentrouting of messages directly to nearby copies of an objector service using only localized resources. Tapestry supportsa generic decentralized object location and routing (DOLR)applications programming interface (API) that uses a self-re-pairing, soft-state-based routing layer. This paper presentsthe Tapestry architecture, algorithms, and implementation. Italso explores the behavior of Tapestry deployment on a globaltestbed under a variety of conditions.The second paper in this area, “Design and Evaluation of aDistributed Scalable Content Discovery System,” focuses on0733-8716/04$20.00 © 2004 IEEE2 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 1, JANUARY 2004a content discovery system (CDS), which is used by nodes inthe system to discover contents published by some other nodesin the system. The main challenge in the design of a CDS isto achieve both rich functionality and scalability. This paperpresents a distributed and scalable CDS that uses rendezvouspoints (RPs) that avoid network-wide message flooding at bothregistration and query time. This paper also presents a mecha-nism that uses load balancing matrices (LBMs) to dynamicallybalance both registration and query load in a system in order toimprove the system’s throughput under skewed load.C. MeasurementMeasurement is an important component for operation andmanagement of overlay networks. The next two papers addressthe techniques and infrastructure for conducting measurements.The first paper in this area, “Computing the Unmeasured:An Algebraic Approach to Internet Mapping,”
View Full Document
Unlocking...