Latency-aware Traffic Provisioning for Content Delivery Networks

With the explosion of network data transmission, Content Delivery Networks(CDNs) cache and deliver hundreds of trillions of user requests each day from thousands of servers around the world. The traffic served by CDNs can be partitioned into thousands of domain requests, each with different user access patterns. One critical problem is how to reduce the time between request and response for content as perceived by the end-user. This paper proposes LaMM, an algorithm that provides domain traffic to servers such that end-user latency is minimized. Based on traffic's cache hit rate curve and the eviction time curve data, we also propose a method to calculate the mixed traffic cache hit rate. Using extensive traces from a commercial CDN system, we show that our traffic provisioning schemes can reduce response latency by nearly 10% compared to the best current latency-unaware methods. Further, we show that our algorithm consistently performs well under various settings and environments.