Balancing Revenue and Cache Partition Fairness for CDN Providers

Nowadays, production CDNs hold a variety of content in shared caches, such as videos, software downloads, web pages, etc., which belongs to hundreds of content providers. Manage cache resources as a single un-partitioned cache and cache the most popular content maximizes cache hit ratio and the revenue of CDN providers. However, this causes some CPs' content files with low request rates never cached, which leads to performance degradation. Although the state-of-art works have focused on cache partitioning, a mechanism that can flexibly balance these two conflicting objectives is still missing. To fill this gap, in this paper, we propose a cache partition method to achieve flexible trade-offs between revenue and caching sharing fairness for CDN providers managing cache partition with LRU (least-recently-used) eviction algorithm. More specifically, our method achieves such flexible trade-offs by finding the Optimal Revenue Fairness Trade-off (ORJT) cache partition policies. It proves to be Pareto optimal by maximizing revenue while ensuring fairness is higher than a given threshold. However, a series of complex non-convex optimization problems get involved in obtaining RJT cache partition policies. Thus, we propose a simple but efficient heuristic algorithm called Landslide to eliminate this difficulty. Landslide simulates the natural landslide phenomenon, it converges to the ORJT cache partition policy by modifying an arbitrary feasible cache partition policy iteratively. Finally, experimental results show the proposed method can achieve flexible trade-offs between CDN provider's revenue and cache partition fairness.