Model-Based Design And Analysis Of Cache Hierarchies

The caching of web and video content is a key component of today's Internet. Caches are often organized as hierarchies where one cache can retrieve and store content from other caches. Cache hierarchies have three objectives: decreasing the traffic served by the content provider's origin (called origin offload), decreasing the cost of transporting content over the WAN (called midgress reduction), and providing a faster response time for users downloading the content. While there is a large body of research on individual caches, much less is known about cache hierarchies. To gain insights into cache hierarchies, we develop an analytical model that can be applied to arbitrary cache hierarchies described by a directed acyclic graph. We show that our model predicts hit rates and other metrics for upper layer caches within cache hierarchies significantly better than prior work. We use our model to shed light on fundamental questions of cache hierarchy design such as how to size individual caches and how to store and route content in a cache hierarchy. Using real world CDN traces and implementable cache eviction policies, we provide insights that are applicable to real-world cache hierarchy designs.