A Lightweight and Adaptive Cache Allocation Scheme for Content Delivery Networks

Abstract

Content delivery networks (CDNs) caching systems usually use multi-tenant shared caching due to their operational simplicity. However, this approach often results in interference among applications. Dynamic cache allocation schemes based on miss ratio curve (MRC) could be a good choice except for its high computational overheads and performance fluctuations. In this paper, we propose a lightweight and adaptive cache allocation scheme for CDNs (LACA). Rather than searching near-optimal configurations for each tenant, LACA detects in real time whether any tenants are using cache space inefficiently (named abnormal tenants), and then adjusts space restricted within these abnormal tenants by constructing their local MRCs instead of the global ones. We have deployed LACA in Tencent's CDN system and LACA can reduce the miss ratio by 27.1 % and reduce the average user access latency by 28.5 ms. Compared with the-state-of-the-art schemes, LACA also achieves a higher-accuracy local MRC with marginal overhead.