Deleting files in the Celeste peer-to-peer storage system

Celeste is a robust peer-to-peer object store built on top of a distributed hash table (DHT). Celeste is a working system, developed by Sun Microsystems Laboratories. During the development of Celeste, we faced the challenge of complete object deletion, and moreover, of deleting “files” composed of several different objects. This important problem is not solved by merely deleting meta-data, as there are scenarios in which all file contents must be deleted, e.g., due to a court order. Complete file deletion in a realistic peer-to-peer storage system has not been previously dealt with due to the intricacy of the problem — the system may experience high churn rates, nodes may crash or have intermittent connectivity, and the overlay network may become partitioned at times. We present an algorithm that eventually deletes all file contents, data and meta-data, in the aforementioned complex scenarios. The algorithm is fully functional and has been successfully integrated into Celeste.     

Reliable Distributed Storage

A distributed storage service lets clients abstract a single reliable shared storage device using a collection of possibly unreliable computing units. Algorithms that implement this abstraction offer certain tradeoffs and vary according to dimensions such as complexity, the consistency semantics provided, and the types of failures tolerated.     

Distributed data clustering in sensor networks

Low overhead analysis of large distributed data sets is necessary for current data centers and for future sensor networks. In such systems, each node holds some data value, e.g., a local sensor read, and a concise picture of the global system state needs to be obtained. In resource-constrained environments like sensor networks, this needs to be done without collecting all the data at any location, i.e., in a distributed manner. To this end, we address the distributed clustering problem, in which numerous interconnected nodes compute a clustering of their data, i.e., partition these values into multiple clusters, and describe each cluster concisely. We present a generic algorithm that solves the distributed clustering problem and may be implemented in various topologies, using different clustering types. For example, the generic algorithm can be instantiated to cluster values according to distance, targeting the same problem as the famous k-means clustering algorithm. However, the distance criterion is often not sufficient to provide good clustering results. We present an instantiation of the generic algorithm that describes the values as a Gaussian Mixture (a set of weighted normal distributions), and uses machine learning tools for clustering decisions. Simulations show the robustness, speed and scalability of this algorithm. We prove that any implementation of the generic algorithm converges over any connected topology, clustering criterion and cluster representation, in fully asynchronous settings.     

Threshold of stimulated Brillouin scattering by use of a solar pumped laser

What is to our knowledge the first stimulated Brillouin scattering experiment using a high-power low-gain solar pumped laser is presented. A threshold reflectivity of 0.23% was reached when a peak power of 20.7 kW was used at 7.6 GHz. A cw solar pumped laser was Q-switched with an acousto-optic modulator, and the bandwidth was narrowed with an intracavity etalon. A high polarization ratio (>99.4%) was achieved by use of an intracavity configuration. Higher reflectivity values were limited because of the lack of availability of optical switches.     

Keeping Denial-of-Service Attackers in the Dark

We consider the problem of overcoming (distributed) denial-of-service (DoS) attacks by realistic adversaries that have knowledge of their attack's successfulness, for example, by observing service performance degradation or by eavesdropping on messages or parts thereof. A solution for this problem in a high-speed network environment necessitates lightweight mechanisms for differentiating between valid traffic and the attacker's packets. The main challenge in presenting such a solution is to exploit existing packet-filtering mechanisms in a way that allows fast processing of packets but is complex enough so that the attacker cannot efficiently craft packets that pass the filters. We show a protocol that mitigates DoS attacks by adversaries that can eavesdrop and (with some delay) adapt their attacks accordingly. The protocol uses only available efficient packet-filtering mechanisms based mainly on addresses and port numbers. Our protocol avoids the use of fixed ports and instead performs "pseudorandom port hopping." We model the underlying packet-filtering services and define measures for the capabilities of the adversary and for the success rate of the protocol. Using these, we provide a novel rigorous analysis of the impact of DoS on an end-to-end protocol and show that our protocol provides effective DoS prevention for realistic attack and deployment scenarios.     

Octopus: A fault-tolerant and efficient ad-hoc routing protocol

Mobile ad-hoc networks (MANETs) are failure-prone environments; it is common for mobile wireless nodes to intermittently disconnect from the network, e.g., due to signal blockage. This paper focuses on withstanding such failures in large MANETs: we present Octopus, a fault-tolerant and efficient position-based routing protocol. Fault-tolerance is achieved by employing redundancy, i.e., storing the location of each node at many other nodes, and by keeping frequently refreshed soft state. At the same time, Octopus achieves a low location update overhead by employing a novel aggregation technique, whereby a single packet updates the location of many nodes at many other nodes. Octopus is highly scalable: for a fixed node density, the number of location update packets sent does not grow with the network size. And when the density increases, the overhead drops. Thorough empirical evaluation using the ns2 simulator with up to 675 mobile nodes shows that Octopus achieves excellent fault-tolerance at a modest overhead: when all nodes intermittently disconnect and reconnect, Octopus achieves the same high reliability as when all nodes are constantly up. A preliminary version of this paper appears in Proceedings of the 24th IEEE Symposium on Reliable Distributed Systems (SRDS 2005) October 26–28, 2005, Orlando, Florida. Idit Keidar is a faculty member at the department of Electrical Engineering at the Technion Israel Institute of Technology, and a recipient of the national Alon Fellowship for new faculty members. She holds Ph.D., M.Sc. (summa cum laude), and B.Sc (summa cum laude) degrees from the Hebrew University of Jerusalem. She was a postdoctoral research associate at MIT’s laboratory for Computer Science, where she held post-doctoral fellowships from Rothschild Yad-Hanadiv and NSF CISE. Dr. Keidar has consulted for BBN Technologies (a Verizon Company) in the area of fault-tolerance and intrusion tolerance, and for Microsoft Research in the area of fault-tolerant storage systems. Dr. Keidar’s research focuses on reliability in distributed algorithms and system. She is the academic head of Software Systems Laboratory at the Technion. Dr. Keidar served as a member of the Steering Committee of the ACM Symposium on Principles of Distributed Computing (PODC), has served on numerous program committees of leading conferences in the area of distributed and parallel computing, has twice served as a vice-chair for the IEEE International Conference on Distributed Computing Systems (ICDCS), and once served as a vice-chair for Euro-Par. Yoav Barel is a Senior Business Line Manager at Sun Microsystems. He holds a B.Sc from the Technion—Israel Institute of Technology. Mr. Barel works with wireless carriers world wide and assists them in deploying innovative services based on cutting edge Java technologies. Roie Melamed is a research staff member at IBM Haifa Research Laboratory. He holds Ph.D. and B.A. (cum laude) degrees from the Technion—Israel Institute of Technology. Dr. Melamed’s research focuses on reliability in distributed systems.     

MaGMA: mobility and group management architecture for real‐time collaborative applications

We introduce MaGMA, a mobility and group management architecture, enabling real‐time collaborative group applications such as push‐to‐talk (PTT) for mobile users. MaGMA provides, for the first time, a comprehensive and scalable solution for group management, seamless mobility, and quality‐of‐service (QoS). MaGMA is a distributed IP‐based architecture consisting of an overlay server network deployed as part of the service infrastructure. MaGMA's architecture consists of a collection of mobile group managers (MGMs), which manage group membership and may also implement a multicast overlay for data delivery. The architecture is very flexible, and can co‐exist with current as well as emerging wireless network technologies. We see such services as essential components in beyond‐3G (B3G) networks. We propose two group management approaches in the context of MaGMA. We devise protocols for both approaches, evaluate both solutions using simulations, and validate the results through mathematical analysis. Finally, we present a proof‐of‐concept prototype implementation. Copyright © 2005 John Wiley & Sons, Ltd.     

EquiCast: Scalable multicast with selfish users

Peer-to-peer (P2P) networks suffer from the problem of “freeloaders”, i.e., users who consume resources without contributing anything in return. In this paper, we tackle this problem taking a game theoretic perspective by modeling the system as a non-cooperative game. We introduce EquiCast, a wide-area P2P multicast protocol for large groups of selfish nodes. EquiCast is the first P2P multicast protocol that is formally proven to enforce cooperation in selfish environments. Additionally, we prove that EquiCast incurs a low constant load on each user.     

A simple proof of the uniform consensus synchronous lower bound

We give a simple and intuitive proof of an f+2 round lower bound for uniform consensus. That is, we show that for every uniform consensus algorithm tolerating t failures, and for every f?t−2, there is an execution with f failures that requires f+2 rounds.     

LiMoSense: live monitoring in dynamic sensor networks

We present LiMoSense, a fault-tolerant live monitoring algorithm for dynamic sensor networks. This is the first asynchronous robust average aggregation algorithm that performs live monitoring, i.e., it constantly obtains a timely and accurate picture of dynamically changing data. LiMoSense uses gossip to dynamically track and aggregate a large collection of ever-changing sensor reads. It overcomes message loss, node failures and recoveries, and dynamic network topology changes. The algorithm uses a novel technique to bound variable size. We present the algorithm and formally prove its correctness. We use simulations to illustrate its ability to quickly react to changes of both the network topology and the sensor reads, and to provide accurate information.