Static site for GPUCache, the shared KV Cache at the G3.5 tier, with a CPU/TCP implementation and a DPU/GDS target architecture.
GPUCache gives a GPU cluster a shared KV Cache hot tier. It is written in Rust, licensed under Apache 2.0, and in development. RustFS is the G4 capacity tier.
| Language | Path |
|---|---|
English (x-default) |
/ |
| 日本語 | /ja/ |
| Deutsch | /de/ |
| Français | /fr/ |
| 中文 | /zh/ |
Each language has a docs section: architecture, tiering, roadmap, quick start, and node scaling. Pages include hreflang alternates for en, ja, de, fr, zh, and x-default.
index.htmlandja/,de/,fr/,zh/— landing pagesdocs/— English docs; other languages live under their prefixstyles.css— RustFS design tokens and layoutscript.js— menu and code-copy buttondesign-tokens.md— token roles and copy constraintsfavicon.svg— site icon
From the repository root:
python3 -m http.server 8080Then open http://localhost:8080/. Root-relative links need a server; opening a file directly will not resolve them.
The scaling guide describes operator-coordinated membership for 1, 3 and 10 shards. Its executable source is crates/transport/examples/scaling_smoke.rs in the GPUCache repository. That example uses independent TCP servers in one localhost process; it is not a physical multi-host or performance qualification. Routes do not fence remote stores, and new stores must be prepared above the previous epoch maximum.
Real RDMA, GDS, BlueField-3, live RustFS and real engine export/import remain unqualified. Keep CPU/mock/fake-block evidence separate from target architecture claims. Run python3 scripts/check_site.py before opening a PR. There is no site deployment step in that command.
Reviewed implementation revision: af51629. The scaling guide links to immutable copies of the matching runbook and example.