RevOPT: An LSTM-based Efficient Caching Strategy for CDN

In order to face the rise in data consumption and network congestion, caching structures like Content Delivery Networks (CDNs) are being more and more used and integrated into the network infrastructure. Knowing that the capacities of caching resources are most often limited due to their large operational cost, it has become very important that these entities are managed efficiently. Especially, at the caching operations level, the question that arises is what content should be cached or evicted from the cache when it becomes full. Having these in mind, we introduce a lightweight Artificial Intelligence-based caching scheme called Reversed OPT (RevOPT). In our proposal, we use a Long Short-Term Memory (LSTM) encoder-decoder model to learn future requests patterns from the past and exploit its outcome with a Counting Bloom Filter (CBF) structure to manage efficiently the caching decisions and to keep in the cache only contents expected to be reused in the near future. The conducted simulations show promising results of RevOPT in terms of the cache hit ratio compared to existing caching algorithms.