Exploring Non-typical Memcache Architectures for Decreased Latency and Distributed Network Usage

Memcache is a distributed in-memory data store designed to reduce database load for web applications by caching frequently used data across multiple machines. While memcache already offers excellent performance, we explore how data-locality can increase performance under certain environments and workloads. We build an analytical model, then compare typical configurations to a handful of proposed schemes which promise to reduce latency and/or reduce core network traffic. An example web application, MediaWiki, is profiled and shown to benefit from alternate Memcache configurations when used in a large web farm. We develop metrics to predict any web application’s performance under our schemes.