This paper addresses the problem of churn---the continuous process of node arrival and departure---in distributed hash tables (DHTs). We argue that DHTs should perform lookups quickly and consistently under churn rates at least as high as those observed in deployed P2P systems such as Kazaa. We then show through experiments on an emulated network that current DHT implementations cannot handle such churn rates. Next, we identify and explore three factors affecting DHT performance under churn: reactive versus periodic failure recovery, message timeout calculation, and proximity neighbor selection. We work in the context of a mature DHT implementation called Bamboo, using the ModelNet network emulator, which models in-network queuing, cross-traffic, and packet loss. These factors are typically missing in earlier simulationbased DHT studies, and we show that careful attention to them in Bamboo's design allows it to function effectively at churn rates at or higher than that observed in P2P file-sharing applications, while using lower maintenance bandwidth than other DHT implementations.

Diagnosing faults in the Internet is arduous and time-consuming, in part because the network is composed of diverse components spread across many administrative domains. We consider an extreme form of this problem: can end users, with no special privileges, identify and pinpoint faults inside the network that degrade the performance of their applications? To answer this question, we present both an architecture for user-level Internet path diagnosis and a practical tool to diagnose paths in the current Internet. Our architecture requires only a small amount of network support, yet it is nearly as complete as analyzing a packet trace collected at all routers along the path. Our tool, tulip, diagnoses reordering, loss and significant queuing events by leveraging well deployed but little exploited router features that approximate our architecture. Tulip can locate points of reordering and loss to within three hops and queuing to within four hops on most paths that we measured. This granularity is comparable to that of a hypothetical network tomography tool that uses 65 diverse hosts to localize faults on a given path. We conclude by proposing several simple changes to the Internet to further improve its diagnostic capabilities.

Peer-to-peer and overlay networks allow routing to be controlled at the application layer. Consider several independent overlay flows, each with a set of available overlay routes to send their data on. If they each select the route with the most available bandwidth, i.e., they are "greedy", a significant degree of instability could result, leading to degraded performance. We investigate this possibility, and a wide variety of factors that affect routing performance, by simulations. We find that some measure of "restraint" is crucial for obtaining acceptable performance of route selection in such scenarios. Specifically, we investigate three forms of restraint - randomization of route selection, utilizing an appropriate hysteresis threshold when switching routes, and increasing the time intervals between route-change considerations. Our results

The Internet has successfully provided a set of protocols for global deployment and re-liable transportation services. With the tremendous growth of the Internet, emerging applications have been continuously demanding the Internet infrastructure to provide new value-added services. In this dissertation, we first focus on the issue of proving the most commonly used value-added Internet service, namely, multicasting. The second part of this dissertation focuses on overlay networks, which is a feasible and flexible approach to quickly deploy value-added services on top of the existing Internet infrastructure. Most multicast applications are inherently QoS (quality of service) sensitive and desire QoS support from the underlying network. Our focus is to design and develop efficient protocols to support QoS-aware and failure-resilient multicasting services in various environments: (1) Integrated services (IntServ) environment: we propose an IP-based QoS-aware multicasting protocol called QMIS based on per-flow reservation. On top of a novel bounded flooding technique, QMIS can increase the

This paper describes a distributed infrastructure for querying network graphs with recursive queries. We argue that recursive queries have great practical value as a declarative interface to multi-hop networks. To query these networks in a distributed fashion, we describe the processing of recursive queries using PIER, a P2P relational query processor that utilizes distributed hash tables (DHTs). We focus on studying a set of commonlyused recursive queries based on the transitive closure query. We demonstrate that different query processing techniques will lead to tradeoffs in latency, work sharing and communication overheads. Our experimental results also show that selecting the best execution strategy based on the query workload and graph topology can lead to significant reduction in communication overhead and latency.

Peer-to-peer and overlay networks allow routing to be controlled at the application layer. Consider several independent overlay flows, each with a set of available overlay routes to send their data on. If they each select the route with the most available bandwidth, i.e., they are "greedy", a significant degree of instability could result, leading to degraded performance. We investigate this possibility, and a wide variety of factors that affect routing performance, by simulations. We find that some measure of "restraint" is crucial for obtaining acceptable performance of route selection in such scenarios. Specifically, we investigate three forms of restraint - randomization of route selection, utilizing an appropriate hysteresis threshold when switching routes, and increasing the time intervals between route-change considerations.

This paper addresses the problem of churn---the continuous process of node arrival and departure---in distributed hash tables (DHTs). We demonstrate through experiment that existing DHT implementations break down at churn levels observed in deployed peer-to-peer systems, contrary to simulation-based results. We present Bamboo, a DHT that handles high levels of churn, and discuss the manner in which it does so. We show that Bamboo is able to function effectively for median node session times as short as 1.4 minutes, while using less than 900 bytes/s/node of maintenance bandwidth in a 1000-node system. This churn rate is faster than that observed in real file-sharing systems such as Gnutella, Kazaa, Napster, and Overnet. Since Bamboo's bandwidth usage scales logarithmically in the number of nodes, we expect this cost to remain within the reach of dialup modems even for very large systems. Moreover, in simulated networks without churn, Bamboo achieves lookup performance comparable with Pastry, an existing DHT with a similar structure.