SIEVE is a highly effective and simple eviction algorithm that determines which items to remove when caching new ones. It outperforms other state-of-the-art algorithms and has a lower miss ratio on a majority of cache systems. What makes SIEVE interesting is that it is a small change added to a basic FIFO queue, making it efficient and reducing the synchronization needed in multi-tenant settings. The reason behind SIEVE’s late discovery might be due to its lack of scan resistance, which has been a focus in caching research. The article also explores a variant of SIEVE called SIEVE-k, which introduces a counter to prioritize popular objects during evictions. Although SIEVE-k shows promising results in certain scenarios, it is not consistently better than SIEVE in real-world use cases. Overall, SIEVE offers a valuable contribution to cache systems.
https://brooker.co.za/blog/2023/12/15/sieve.html