next up previous
Next: Setup Up: Beehive: O(1) Lookup Performance Previous: Implementation


Evaluation

In this section, we evaluate the performance costs and benefits of the Beehive replication framework. We examine Beehive's performance in the context of a DNS system and show that Beehive can robustly and efficiently achieve its targeted lookup performance. We also show that Beehive can adapt to sudden, drastic changes in the popularity of objects as well as global shifts in the parameter of the query distribution, and continue to provide good lookup performance.

We compare the performance of Beehive with that of pure Pastry and Pastry enhanced by passive caching. By passive caching, we mean caching objects along all nodes on the query path, similar to the scheme proposed in [23]. We impose no restrictions on the size of the cache used in passive caching. We follow the DNS cache model to handle mutable objects, and associate a time to live with each object. Objects are removed from the cache upon expiration of the time to live.



Subsections

beehive-l@cs.cornell.edu 2004-02-11