Abstract
Designing a cache front end for cloud storage calls into question of the effectiveness of the popular LRU cache eviction policy versus the FIFO heuristic. Several past works have considered this question and commonly stipulated that while FIFO is much easier to implement, the improved hit ratio of LRU outweighs this. Two main trends call for a re-evaluation: the very large scales of cloud storage which makes managing cache metadata in RAM infeasible, and new workloads which possess different characteristics. We model the overall cost of running LRU and FIFO in a very large scale cache and evaluate this cost using traces taken from real world object store system. While there are lots of traces for file and block storage, to date, there has been a dearth of traces for Object Storage. To address this gap, we collected object storage traces from the IBM Public Cloud. IBM has anonymized these traces and made them available to the public at the SNIA IOTTA repository.