Towards Practical Virtual Server-Based Load Balancing for Distributed Hash Tables

Abstract

Current virtual server-based load balancing schemes for DHT have been shown to be able to achieve excellent load balancing effectiveness. However, they face two important issues. They suffer from problems of incurring extremely high overheads, and inducing severe inconsistency in DHT routing state. We present two fundamental components, virtual server management and active stabilization, whose inclusion into these schemes essentially eliminates these problems. As a result, these schemes not only incur overheads comparable to non-virtual server-based systems, but also achieve better query performance.