CoDoNS Cooperative Domain Name System		CobWeb Content Distribution Network		CorONA Cornell Online News Aggregator

Beehive is a high performance peer-to-peer overlay. It uses structured replication to provide O(1) lookup latencies on overlays. The main insight behind Beehive is that replication can be expressed as a principled optimization problem that takes into account the tradeoff between replication costs and the resulting performance of the system. This analytical formulation of the problem enables Beehive to achieve strong guarantees on the expected behavior of the system.

Beehive achieves high performnce by replicating objects. Done naively or haphazardly, replication is unlikely to achieve high performance, especially in the presence of heavy-tailed popularity distribution commonly encountered in many settings including the web. Instead,Beehive performs an informed tradeoff,mathematically capturing the costs and performing just the minimal amount of replication needed to achieve a targeted level of lookup performance. The unique solution strategy used in Beehive enables the system to provide assurance that a certain fraction of queries will be satisfied (i.e. will be cache hits) at the first hop, in essence shipping content to clients even before they have asked for it.

Beehive is powered by a set of low overhead distributed algorithms, which solve optimization problems capturing the performance-overhead tradeoffs. Beehive uses analytical closed-form formulas for determining optimal solutions for Zipf query distributions, and a fast and accurate numerical technique for any query distribution. The numerical technique enables Beehive to achieve substantial reduction in replication overhead by incorporating the wide variance in sizes, and update rates of objects.

Beehive complements the high scalability and failure resilience of structured overlays by providing the following critical properties required to support performance-demanding infrastructure services.

• High Performance: Beehive provides constant lookup performance on average with minimal overhead in storage, network bandwidth, and load.
• Fast Updates: Beehive disseminates updates proactively to all nodes with replicas of the object. It does not use weak consistency mechanisms, such as expiry times (TTL), to handle mutable objects.
• Adaptability: Beehive quickly detects and reacts to large scale upheavals in popularity of objects, which arise during flash crowds and denial of service attacks.

We have used Beehive to build three infrastructure services: a replacement for the domain name service called CoDoNS, an open-access content distribution network called CobWeb, and a publish-subscribe system called CorONA for disseminating Web MicroNews. All these services are currently deployed on Planet-Lab.

Beehive	Overview of Beehive.
CoDoNS	A new name service for the Internet.
CobWeb	An open access content distribution network.
CorONA	A high-performance publish-subscribe system for Web micronews (RSS).
Structured Replication	How Beehive provides high performance, DoS-resilience, and fast update dissemination.
Related Work	Pointers to work on high-performance peer-to-peer systems.
People	Project members.
Papers	Papers and Talks on Beehive and related projects.
Press	News articles on Beehive and related projects.
Support	Project support and funding.

Computer Science Department
Cornell University