ABCS - Adaptive bandit cache selection

select method store Multi-Armed Bandit

Disclaimer

It is experimental solution. Because choosing an algorithm during program execution multiplies memory consumption

Problem

I would like to make an algorithm based on statistical data so that it independently chooses the optimal storage method during operation. Because choosing a specific algorithm for specific tasks is a separate research work

Input

In create cache

• Start algorithm
• Count storage elements
• We won't stop business logic for restructed cache

In runtime

• Hitrate into start algorithm

Questions

• When we need migrate from start algorithm to new?

Idea

TODO

Usage

TODO

Supported cached methods

• Random
• LRU by default
• LFU
• 2Q
• ARC

Reference

TODO

Implementarions

• LRU, ARC, 2Q golang-lru
• LFUDA lfuda-go
• Multi-Armed Bandit(MAB) go-bandit