Search results

Resource discovery in peer‐to‐peer (P2P) systems have been extensively studied. Unfortunately, most of the systems studied are not designed to take advantage of the heterogeneity in peer nodes. In this paper, we propose a novel P2P overlay called RATTAN, which serves as an underlay of a Gnutella‐like network. RATTAN exploits the heterogeneity of peer nodes by structuring capable nodes as the core of the overlay. Using a tree‐like structure, RATTAN can maximize the search scope with a minimal number of query messages. We evaluate RATTAN with simulation. The experiments show the following interesting results. First, RATTAN is robust by exploiting redundant overlay links. Second, the maximum bandwidth demand for processing the protocol of a single RATTAN overlay is nearly 1M bits/sec. However, around 80% of the nodes merely take 66 bits/sec. One implication is that we can use a small number of relatively capable peers (e.g., stable machines with a 100M bits/sec network interface) to process the 1M bits/sec protocol overhead and serve other peers that only need to spend 66 bits/sec for processing protocol overhead.

The purpose of this paper is to propose MUREX, a mutable replica control scheme, to keep one‐copy equivalence for synchronous replication in structured peer‐to‐peer (P2P) storage systems.

For synchronous replication in P2P networks, it is proper to adopt crash‐recovery as the fault model; that is, nodes are fail‐stop and can recover and rejoin the system after synchronizing their states with other active nodes. In addition to the state synchronization problem, the paper identifies other two problems to solve for synchronous replication in P2P storage systems. They are the replica acquisition and the replica migration problems.

On the basis of multi‐column read/write quorums, MUREX conquers the problems by the replica pointer, the on‐demand replica regeneration, and the leased lock techniques.

The paper proves the correctness of MUREX, analyzes and also simulates it in terms of communication cost and operation success rate.

A primary task of wireless sensor networks is to measure environmental conditions. In most applications, a sink node is responsible for collecting data from the sensors through multihop communications. The communication pattern is called convergecast. However, radio congestion around the sink can easily become a bottleneck for the convergecast. The purpose of this paper is to consider both scheduling algorithms and routing structures to improve the throughput of convergecast.

The paper addresses the issue from two perspectives. First by considering the transition scheduling that reduces radio interference to perform convergecast efficiently. Second, by studying the effects of routing structures on convergecast. A routing algorithm, called disjoint‐strip routing, is proposed as an alternative to existing shortest‐path routing.

The paper shows that constructing a shortest‐length conflict‐free schedule is equivalent to finding a minimal vertex coloring. To solve the scheduling problem, a virtual‐node expansion is proposed to handle relay operations and then coloring algorithms are utilized. Regarding the routing structures, a disjoint‐strip algorithm is proposed to leverage possible parallel transmissions. Proposed algorithms are evaluated through simulations.

This paper separates the problem for optimizing data‐collection throughput into two stages: constructing a routing structure on a given deployment; and scheduling the activation time of each link. Determining routing topologies and communication schedules for optimal throughput are shown to be hard, so heuristics are applied in both stages. VNE is proposed, which makes traffic information visible to coloring algorithms. The advantage of VNE is verified through simulations. VNE can be applied to any coloring algorithm and any deterministic traffic pattern. It is shown that routing structures set a limit on the performance of scheduling algorithms. There are two possible ways in routing algorithms to improve convergecast throughput: first, by reducing the total number of transmissions during data collection; second, by transferring data in parallel. The shortest‐path routing addresses the first point while DS addresses the second one. As expected, when the deployments are even and balanced, minimizing the number of transmissions is more effective than parallelizing them. On the other hand, when the deployments are unbalanced and conflicts are not strict, parallel transmissions can improve the throughput.