Scalable strong consistency for web applications

Abstract

Web application workloads are often characterized by a large number of unique read requests and a significant fraction of write requests. Hosting these applications drives the need for the next generation CDN architecture that does more than caching the results of Web applications but replicates both the application code and its underlying data. We propose the design of a system that guarantees strong consistency for Web applications with high scalability. The proposed system is based on partial replication, where data units are replicated only to servers that access them often. This reduces the consistency overhead as updates are sent to a reduced number of servers. The novelty of our system is that the proposed partial replication is performed by the system automatically by analyzing the system's access patterns periodically. We explore the design space of this system, find the key issues that need to be addressed to build it and propose solutions to solve them. We further show that the proposed algorithms offer significant performance gains compared to existing solutions for a wide range of Web access patterns.