Use simple `std::mutex` in the cache class

[morinim]

The cache hash table currently uses shared locks (std::shared_mutex) to enable concurrent reads and exclusive writes. However, internal tests conducted on standard workstations indicate that the overhead of std::shared_mutex may outweigh its benefits, particularly when compared to std::mutex. This appears to be due to the short critical section in the find member function, which might not justify the complexity of read-write locks.

Observations

• On systems with a limited number of CPUs or hardware threads, the contention for std::shared_mutex may introduce significant overhead, reducing the performance advantage of concurrent reads.
• On systems with a greater number of CPUs or hardware threads, the benefits of std::shared_mutex may become more apparent, as contention is distributed across more resources.

Cross-Platform Considerations

• Differences between operating systems, particularly Linux and Windows, may lead to varying performance characteristics for std::shared_mutex due to differences in underlying implementations.
• Further testing is required to evaluate these differences and ensure consistent performance across platforms.

Recommendations

Benchmarking

• Perform detailed benchmarks comparing std::shared_mutex and std::mutex on systems with varying CPU and thread counts.
• Evaluate performance differences across Linux and Windows platforms.

Simplification

Consider replacing std::shared_mutex with std::mutex if benchmarks show consistent performance improvements and the benefits of concurrent reads are negligible.

References:

• https://news.ycombinator.com/item?id=39582827
• https://petsta.net/blog/2022-09-30-rwmutex/

[morinim]

The simplicity of std::mutex can also help avoid subtle bugs or complexity in debugging lock contention.