Adaptive Buffer Replacement Policies for Multi-Tenant Cloud Services

Cloud services such as multi-tenant content delivery networks (CDN) have become a trend because they can offer personalized business support for tenants. However, the business modes of different tenants are usually different. Thus, it is not appropriate to use a single buffering strategy on cloud servers because it cannot adapt to various access patterns of tenants. In this paper, we propose to develop adaptive buffer management for multi-tenant cloud services. Particularly, we present an adaptive buffer replacement policy called ARP (Adaptive Replacement Policy) for each tenant. ARP can automatically choose a recency-based or a frequency-based replacement policy according to the workload characteristics. Further, to solve the problem that ARP needs to manually set a parameter $p$ , we present an optimization for ARP, called ARP+, to automatically adjust the value of parameter $p$ according to the workload characteristics. We conduct experiments on multiple tailor-made workloads. The results show that both ARP and ARP+ outperform conventional buffering policies, including LRU, LFU, LIRS, and LRU-2. In addition, ARP+ can automatically adapt to the change of access patterns, which is more suitable for multi-tenant cloud scenarios.