LRU Cache Benchmark

In this project, we benchmark the performance of four different implementations of a LRU-type cache:

• Our implementation using the "lock-free locks" based DList and Hashmap from cmuparlay/flock (C++)
• Our fine-grained locking implementation built on Intel TBB Hashmap and Scoped Locks (C++)
• Caffeine Cache (Java)
• Guava Cache (Java)

Benchmarking Setup

• Google Benchmark for the locking implementation
• cmuparlay for the lock-free implementation
• Java Microbenchmark Harness (JMH) for Caffeine and Guava

Variables Tested

• Zipfian parameter for get/put key distribution (0.01, 0.75, 0.9, 0.99)
  • dirtyzipf for C++
  • YCSB for Java
• Number of threads used during benchmarking (1, 4, 16, 64, 128 (overclock))
• Cache Size (100, 10000, 100000, 1000000)

Dependencies

• CMake ≥ 3.13
• C++17
• OpenJDK 24.0.1
• Maven 3.9.9
• Google Benchmark
• JMH
• TBB
• cmuparlay/flock
• cmuparlay/parlaylib
• dirtyzipf
• YCSB
• Caffeine
• Guava