https://github.com/bootjp/adaptsize-go

Go AdaptSize cache: size-aware admission exp(-size/c) + LRU eviction with background tuner maximizing OHR, for KVS front-ends.
https://github.com/bootjp/adaptsize-go

adaptsize admission-control cache caching kvs lru size-aware-caching

Last synced: 2 months ago
JSON representation

Go AdaptSize cache: size-aware admission exp(-size/c) + LRU eviction with background tuner maximizing OHR, for KVS front-ends.

• Host: GitHub
• URL: https://github.com/bootjp/adaptsize-go
• Owner: bootjp
• Created: 2025-11-03T10:55:26.000Z (7 months ago)
• Default Branch: main
• Last Pushed: 2025-12-24T05:38:09.000Z (6 months ago)
• Last Synced: 2025-12-25T18:52:28.123Z (6 months ago)
• Topics: adaptsize, admission-control, cache, caching, kvs, lru, size-aware-caching
• Language: Go
• Homepage:
• Size: 446 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          

# AdaptSize cache for Go KVS

- Admission: `P(admit) = exp(-size/c)`

- Eviction: LRU

- Background tuning: maximize OHR by grid-searching `c`; for each `c`, solve `μ` s.t. `Σ P_in(i)*s_i = K` via binary search.

## Usage

```go

cache := adaptsize.New(adaptsize.Options{

    CapacityBytes: 1<<30, // 1 GiB

    WindowN:       250_000,

})

defer cache.Close()

// For every request, record it and decide admission on misses.

req := adaptsize.Request{Key: "k1", SizeBytes: 1234, Hit: false} // hit comes from your cache

admit := cache.Request(req)

if !req.Hit && admit {

    // insert into your cache implementation

}

c := cache.ParameterC()

_ = c

```

## Acknowledgments and References

This library is a reimplementation based on the formulas and design principles of **AdaptSize**, a paper by Akamai researchers published at NSDI 2017. We acknowledge the prior work of the authors and their institutions. This is an independent third-party implementation and is not affiliated with the paper’s authors, their institutions, or the conference.

**References**

- AdaptSize, Proceedings of the 14th USENIX Symposium on Networked Systems Design and Implementation (NSDI), 2017. Akamai Technologies.  

  Core elements in this implementation: size-aware admission `exp(-size/c)`, a closed-form in-cache probability derived from the theorem, solving `μ` from the capacity constraint, and a global search that maximizes OHR (Object Hit Ratio).

**Notes**

- The API and parameter tuning implemented here follow the paper’s model; we adopt `exp(-size/c)` to match the paper’s notation.

- This code is not the paper’s official implementation. Algorithmic accuracy and applicability depend on the workload.