Cache management in content delivery networks using the metadata of online social networks

Abstract

The number of requests on content delivery networks (CDN) originating from the online social networks (OSN) by sharing the content weblink increases according to recorded data. The sequence of the OSN originated requests shows temporal burstiness with a typical interval shorter than the ordinary requests. We consider CDN and OSN as a multilayer network and exploit the average spreading power of each user in the OSN to predict the temporal pattern of the corresponding consecutive social requests that may originate from this user to improve the underlying cache management mechanism. The traditional least recently used (LRU) content replacement algorithm uses the statistical popularity of contents to increase the cache’s hit ratio. We propose LRU-Social, which defers the eviction of social requests for a specific amount of time to take advantage of the possible burstiness in the underlying interval without missing the popular contents’ hits. We model the content link sharing by the susceptible–infected–recovered (SIR) spreading process in the underlying OSN to compute the user spreading power. We provide numerical studies for synthetic streams consisting of ordinary requests that follow Zipf’s popularity model and social requests to justify the effectiveness of the LRU-Social compared to the LRU.