Replication of data and services is a basic building block in the design of distributed systems over wide-area networks (WANs). WAN replication is a complex multi-dimensional problem that seeks to optimize metrics such as availability, response time, and freshness of content while respecting constraints such as network connectivity, network bandwidth, storage space, computing bandwidth, and consistency requirements imposed by the service.

In this talk, I present a self-tuning architecture for WAN replication systems that simplifies the construction of replicated services, improves performance by automatically adapting to available resources, and increases robustness to overload situations. These properties are achieved by making aggressive speculative replication possible in a self-tuning manner through mechanisms such as TCP Nice, a new end-to-end transport protocol for background traffic. On the other hand, current replication systems, based on more than a decade of research on speculative replication forr WANs, are unable to effectively avail of such benefits as they are mired in the complexity of manually tuned architectures that are risk-prone and inefficient. I demonstrate the effectiveness of our architecture through a case study exploring the costs and benefits of a Web prefetching system.

Bio

Arun Venkataramani is Ph.D. candidate at the University of Texas at Austin where he is advised by Mike Dahlin. His research interests span distributed systems, networking, security, fault tolerance, and algorithms. He earned his B.S. from the Indian Institute of Technology Bombay and M.S. from the University of Texas at Austin.