Add Sieve implementation and Integrate Sieve into Cachebench

[Helen0721]

Sieve is a FIFO-based eviction algorithm recognized for its throughput and scalability. We’ve integrated Sieve as a new MMContainer option. During testing with key-value traces and synthetic traces using cachebench, Sieve demonstrated a comparable or higher hit ratio while achieving higher scalability compared to other algorithms.

Testing Constraints:
Due to hardware limitations, we capped the amplification factor at 10 and set the cache size to 40GB (approximately 10% of the total cache footprint) and 10GB. The results are shown here:
40GB_t1_lru,lru2q,sieve_243.pdf
40GB_t20_lru,lru2q,sieve_195.pdf
10GB_t1_lru,lru2q,sieve_500.pdf
10GB_t1_lru,lru2q,sieve_195.pdf

Throughput Considerations:
Please note that throughput during testing was constrained by the CSV generator. In synthetic trace tests, Sieve achieves higher get rates and hit ratios:
8192MB_t48_lru,lru2q,sieve.pdf

How to Use Sieve:
To utilize Sieve, specify it as an allocator using cachelib::SieveAllocator .

[facebook-github-bot]

@pbhandar2 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.