Fenrir: Practical Algorithms for Cache Partitioning

Caches are ubiquitous systems that support the modern web. They are crucial to cope with delays, bandwidth and high energy use of accessing ever-growing data from backend databases. Today, almost every web access passes through some level of cache, starting with the browser cache found within the computer, through a series of caching servers until the data desired is either found in memory on one of the cache servers or is served up from the backend database. In this project, we investigate the problem of cache partitioning. Cache partitioning concerns the following question: given a number of tenants (think corporate customers) on a single server, how much cache space should each tenant be allotted? Moreover, given a set of tenants and a set of servers, where should the data for each tenant be placed so as to maximize each of their hit rates