An experimental TypeScript implementation of the SIEVE Cache algorithm — a modern cache replacement policy designed in 2023. This project demonstrates how SIEVE works using Bun as the runtime and @axonlabs/core as the backend library for showcasing results.

SIEVE is a new cache algorithm created with the goal of being:

• ✅ Simpler than LRU (Least Recently Used)
• 🚀 Higher performing and more scalable
• 🎯 Improved hit rate and practical usability

Instead of using a tail pointer like LRU, SIEVE uses a "hand" pointer that walks through a Key-Value Doubly Linked Queue to decide which node to evict.

You can read full paper here: https://junchengyang.com/publication/nsdi24-SIEVE.pdf

Each cache entry (node) has a visited flag (0/1 or false/true):

1. Cache hit → visited = true

   • Means the data was used at least twice.
   • When the hand reaches it, visited will be reset to false, and the hand moves forward.

2. Cache miss (insert new node)

   • The hand traverses the queue.
   • If it finds a node with visited = false, that node gets evicted.
   • The new node is added to the head of the queue.

This mechanism makes SIEVE faster, leaner, and fairer compared to LRU.

• Language: TypeScript
• Runtime: Bun
• Backend: @axonlabs/core

# clone the repo
git clone https://github.com/Mr-MKZ/Sieve-Cache.git
cd Sieve-Cache

# install dependencies
bun install

import { SieveCache } from "./Sieve.ts";

// Create a cache with capacity 3
const cache = new SieveCache<number>(3);

cache.set("a", 1);
cache.set("b", 2);
cache.set("c", 3);

console.log(cache.get("a")); // hit → 1
cache.set("d", 4);           // eviction triggered by SIEVE

Run with Bun:

bun run src/index.ts
or
bun start

This project includes integration with @axonlabs/core to benchmark and observe SIEVE behavior in a backend context.

Sample Usage

MIT © 2025 — Built with ❤️ using TypeScript + Bun + AxonJs