Hydra: Cloud Computing with Many Heads

Hydra lets idle laptops, desktops, and single-board computers cooperate on embarrassingly-parallel work.
The current prototype pairs a Rust scheduler + worker binary with a Python (Flask) web front-end.

Project status: functional PoC
– ✅ Dynamic chunking for Monte-Carlo π
– ✅ Distributed Mandelbrot rendering (static row chunks)
– ✅ Automatic retry / chunk reassignment
– ⚠️ No secure enclaves yet • No token rewards yet • No carbon-aware dispatch yet

A longer technical write-up (architecture, benchmarks, carbon analysis) is in written_final_report.pdf

---

Live demo

https://hydracompute.com – runs the same code in this repo behind an ALB with a public TLS cert.

---

Quick architecture glance

• Scheduler (scheduler/): keeps per-job state, sizes chunks to ~2 s per worker, reassigns slow or lost chunks.
• Worker (worker/): polls for work, runs either π simulation or Mandelbrot row, submits results.
• WebApp (webapp/): simple UI + polling; shows progress and renders images/charts in real-time.

---

Feature highlights

Area	What works today	Notes
Monte-Carlo π	Adaptive chunk sizes per worker	~8 × speed-up on 8 heterogeneous nodes
Mandelbrot	One row = one chunk	Dynamic chunking planned
Fault tolerance	Over-due chunks auto-requeued; workers can disappear	Parameter = 40 × average chunk time
Transport security	TLS by default (skip verification in local dev)	No SGX/SEV yet – don’t send secrets
Deployment	Cargo binaries or multi-stage Dockerfiles	See below

---

Local quick-start

1 · Prerequisites

• Rust 1.77+ (install with rustup)
• Python 3.10+ (for Flask UI)
• Self-signed cert/key in certs/ (only needed for local HTTPS)

2 · Run the scheduler

cd scheduler
cargo run --release
# listens on https://127.0.0.1:8443

3 · Launch one or more workers

cd worker
cargo run --release -- --local     # --local → trust the self-signed cert
# open new terminals for extra workers

4 · Start the web front-end

cd webapp
python app.py           # http://127.0.0.1:5000

Docker quick-start (optional)

# scheduler
docker build -t hydra-scheduler -f scheduler/Dockerfile .
docker run -p 8443:8443 hydra-scheduler

# worker
docker build -t hydra-worker -f worker/Dockerfile .
docker run hydra-worker --local     # repeat for more instances

# webapp (uses system Python image)
cd webapp
docker build -t hydra-webapp .
docker run -p 5000:5000 hydra-webapp

Roadmap (next 12 months)

• Confidential compute: SGX/SEV support so sensitive data can run on untrusted volunteers
• Grid-aware scheduler: shift chunks toward regions with cleaner electricity
• Adaptive Mandelbrot tiling: variable-height stripes to remove current load imbalance
• Tokenised incentives: optional micro-payments to offset volunteer electricity