The various proposed DHT routing algorithms embody several di#erent underlying routing geometries. These geometries include hypercubes, rings, tree-like structures, and butterfly networks. In this paper we focus on how these basic geometric approaches a#ect the resilience and proximity properties of DHTs. One factor that distinguishes these geometries is the degree of flexibility they provide in the selection of neighbors and routes. Flexibility is an important factor in achieving good static resilience and e#ective proximity neighbor and route selection. Our basic finding is that, despite our initial preference for more complex geometries, the ring geometry allows the greatest flexibility, and hence achieves the best resilience and proximity performance.

Abstract. In recent years, the gossip-based communication model in large-scale distributed systems has become a general paradigm with important applications which include information dissemination, aggregation, overlay topology management and synchronization. At the heart of all of these protocols lies a fundamental distributed abstraction: the peer sampling service. In short, the aim of this service is to provide every node with peers to exchange information with. Analytical studies reveal a high reliability and efficiency of gossip-based protocols, under the (often implicit) assumption that the peers to send gossip messages to are selected uniformly at random from the set of all nodes. In practice—instead of requiring all nodes to know all the peer nodes so that a random sample could be drawn—a scalable and efficient way to implement the peer sampling service is by constructing and maintaining dynamic unstructured overlays through gossiping membership information itself. This paper presents a generic framework to implement reliable and efficient peer sampling services. The framework generalizes existing approaches and makes it easy to introduce new ones. We use this framework to explore and compare several implementations of our abstract scheme. Through extensive experimental analysis, we show that all of them lead to different peer sampling services none of which is uniformly random. This clearly renders traditional theoretical approaches invalid, when the underlying peer sampling service is based on a gossip-based scheme. Our observations also help explain important differences between design choices of peer sampling algorithms, and how these impact the reliability of the corresponding service. 1

Abstract — Over the Internet today, computing and communications environments are significantly more complex and chaotic than classical distributed systems, lacking any centralized organization or hierarchical control. There has been much interest in emerging Peer-to-Peer (P2P) network overlays because they provide a good substrate for creating large-scale data sharing, content distribution and application-level multicast applications. These P2P networks try to provide a long list of features such as: selection of nearby peers, redundant storage, efficient search/location of data items, data permanence or guarantees, hierarchical naming, trust and authentication, and, anonymity. P2P networks potentially offer an efficient routing architecture that is self-organizing, massively scalable, and robust in the wide-area, combining fault tolerance, load balancing and explicit notion of locality. In this paper, we present a survey and comparison of various Structured and Unstructured P2P networks. We categorize the various schemes into these two groups in the design spectrum and discuss the application-level network performance of each group.

Several peer-to-peer networks are based upon randomized graph topologies that permit e#cient greedy routing, e.g., randomized hypercubes, randomized Chord, skip-graphs and constructions based upon small-world percolation networks. In each of these networks, a node has out-degree #(log n), where n denotes the total number of nodes, and greedy routing is known to take O(log n) hops on average. We establish lower-bounds for greedy routing for these networks, and analyze Neighbor-of-Neighbor (NoN)-greedy routing. The idea behind NoN, as the name suggests, is to take a neighbor's neighbors into account for making better routing decisions.

A fundamental challenge in data center networking is how to efficiently interconnect an exponentially increasing number of servers. This paper presents DCell, a novel network structure that has many desirable features for data center networking. DCell is a recursively defined structure, in which a high-level DCell is constructed from many low-level DCells and DCells at the same level are fully connected with one another. DCell scales doubly exponentially as the node degree increases. DCell is fault tolerant since it does not have single point of failure and its distributed fault-tolerant routing protocol performs near shortest-path routing even in the presence of severe link or node failures. DCell also provides higher network capacity than the traditional treebased structure for various types of services. Furthermore, DCell can be incrementally expanded and a partial DCell provides the same appealing features. Results from theoretical analysis, simulations, and experiments show that DCell is a viable interconnection structure for data centers. Categories and Subject Descriptors C.2.1 [Network Architecture and Design]: Network topology, Packet-switching networks

We present a low-cost, decentralized algorithm for ID management in distributed hash tables (DHTs) managed by a dynamic set of hosts. Each host is assigned an ID in the unit interval [0, 1). At any time, the set of IDs splits the interval into disjoint partitions. Hosts do not possess global knowledge of other IDs in the system. The challenge then is to design an efficient decentralized algorithm that maintains roughly equi-sized partitions, in the face of arrivals, departures and changes in the average number of hosts. Our ID management algorithm is the first to enjoy all of the following properties: (a) both arrivals and departures of hosts are handled, (b) departure of a host causes at most one existing host to change its ID, (c) the ratio of the largest to the smallest partition is at most 4, with high probability, and (d) the expected cost per arrival/departure is Θ(R + log n) messages, where n denotes the current number of participants, and R denotes the cost of routing one message in the DHT. In fact, our algorithm is independent of the topology of the overlay network used for routing. Variations of our algorithm diminish the ratio between the largest and the smallest partition to (1 + ɛ), for any ɛ> 0, albeit at the cost of re-assigning the IDs of O ( 1) existing ɛ hosts per arrival/departure. Ours is the first algorithm that allows such fine-tuning. Finally, our ID management algorithm enables (a) estimation of the total number of hosts in the system by making only local measurements, and (b) emulation of a variety of deterministic and randomized families of routing topologies, in a straightforward fashion. Among these families are several networks that require O(log n / log k) routing hops in an n-node network with k links per node.

This paper presents Araneola 1, a scalable reliable application-level multicast system for highly dynamic wide-area environments. Araneola supports multi-point to multi-point reliable communication in a fully distributed manner while incurring constant load (in terms of message and space complexity) on each node. For a tunable parameter k ≥ 3, Araneola constructs and dynamically maintains a basic overlay structure in which each node’s degree is either k or k +1, and roughly 90 % of the nodes have degree k. Empirical evaluation shows that Araneola’s basic overlay achieves three important mathematical properties of k-regular random graphs (i.e., random graphs in which each node has exactly k neighbors) with N nodes: (i) its diameter grows logarithmically with N; (ii) it is generally k-connected; and (iii) it remains highly connected following random removal of linear-size subsets of edges or nodes. The overlay is constructed and maintained at a low cost: each join, leave, or failure is handled locally, and entails the sending of only about 3k messages in total, independent of N. Moreover, this cost decreases as the churn rate increases. The low degree of Araneola’s basic overlay structure allows for allocating plenty of additional bandwidth for specific application needs. In this paper, we give an example for such a need — communicating with nearby nodes; we enhance the basic overlay with additional links chosen according to geographic

One of the key reasons overlay networks are seen as an excellent platform for large scale distributed systems is their resilience in the presence of node failures. This resilience rely on accurate and timely detection of node failures. Despite the prevalent use of keep-alive algorithms in overlay networks to detect node failures, their tradeoffs and the circumstances in which they might best be suited is not well understood. In this paper, we study how the design of various keep-alive approaches affect their performance in node failure detection time, probability of false positive, control overhead, and packet loss rate via analysis, simulation, and implementation. We find that among the class of keep-alive algorithms that share information, the maintenance of backpointer state substantially improves detection time and packet loss rate. The improvement in detection time between baseline and sharing algorithms becomes more pronounced as the size of neighbor set increases. Finally, sharing of information allows a network to tolerate a higher churn rate than baseline.

Identification is an essential building block for many services in distributed information systems. The quality and purpose of identification may differ, but the basic underlying problem is always to bind a set of attributes to an identifier in a unique and deterministic way.

A number of Distributed Hash Table (DHT)-based protocols have been proposed to address the issue of scalability in peer-topeer networks. However, it remains an open question whether there exists a DHT scheme that can achieve the theoretical lower bound of log log n on network diameter when the average routing table size at nodes is no more than log n. In this paper, we present Ulysses, a peer-to-peer network based on the butterfly topology that matches this theoretical lower bound. Compared to existing DHT-based schemes with similar routing table size, Ulysses reduces the network diameter by a factor of log log n, which is 2-4 for typical configurations. This translates into the same amount of reduction on query latency and average traffic per link/node. In addition, Ulysses maintains the same level of robustness in terms of routing in the face of faults and recovering from graceful/ungraceful joins/departures, as provided by existing DHT-based schemes. The protocol is formally verified for its correctness and robustness using techniques from distributed computing. The performance of the protocol has been evaluated using both analysis and simulation.