Morphogenesis PIR

High-throughput Private Information Retrieval for Ethereum light clients.

Current Status

• Full mainnet matrix PIR on H100: 53.0 ms latency, 1.3 TB/s throughput
• Subtree-optimized GPU kernel: 27.4 ms on B200 (2.51 TB/s)
• Storage lookups use 8-byte Cuckoo tags with full 52-byte verification
• Verifiable PIR (sumcheck/binius) gated behind verifiable-pir feature

Performance (Latest)

Metric	Value
GPU scan throughput (H100)	1,300 GB/s
GPU scan latency (68.8 GB)	53.0 ms
GPU scan throughput (B200)	2,510 GB/s
GPU scan latency (68.8 GB)	27.4 ms

Quick Start

# Build with AVX-512 and parallel support
cargo build --release --features avx512,parallel

# Run benchmark (75GB matrix, 3 iterations)
./target/release/bench_scan --rows 78643200 --iterations 3 --warmup-iterations 1 --scan-only --parallel

Development Workflow (MANDATORY)

This project uses mandatory tooling for task tracking and code quality:

backlog.md - Task Management

# View all tasks
backlog task list

# Work on a task
backlog task edit TASK-X -s "In Progress"

# Commit with task reference (auto-linked via git hook)
git commit -m "feat: implement feature

Addresses TASK-X"

roborev - Continuous Code Review

# View latest reviews
roborev list

# Check specific commit
roborev show COMMIT_SHA

# Address findings
roborev address JOB_ID

Git Hook Setup

The .git/hooks/post-commit hook automatically:

1. Triggers roborev review for every commit
2. Links commits to tasks when message includes "Addresses TASK-X"
3. Updates zoekt search index

All commits must reference a task ID and will be automatically reviewed.

Architecture

crates/
├── morphogen-core/     # Core types (DeltaBuffer, EpochSnapshot, GlobalState)
├── morphogen-dpf/      # DPF key trait and implementations
├── morphogen-storage/  # AlignedMatrix, ChunkedMatrix
└── morphogen-server/   # Scan kernel, server, benchmarks

Features

• avx512 - Enable AVX-512 SIMD optimizations
• parallel - Enable multi-threaded chunk processing (rayon)
• profiling - Enable detailed timing instrumentation
• verifiable-pir - Enable sumcheck/binius proof plumbing

Documentation

• Kanban - Project status and tasks (historic)
• Backlog - Active task management (use backlog CLI)
• Protocol & Architecture - v5.0
• Performance - Optimization findings
• Profiling Guide - How to profile
• Cryptography & Core Mechanics - Fused kernel and DPF logic
• Crypto Analysis - Why we use ChaCha8 over AES on GPUs
• Phase 79 Brief - Kernel optimization summary
• Trace Explanation - Sumcheck trace sizing

Experiments

Modal benchmarks and data-prep scripts live in experiments/.

Key Concepts

• Epoch-Based Delta-PIR: Wait-free snapshot isolation for live updates
• Parallel Cuckoo Retrieval: 3 simultaneous DPF queries per request
• Copy-on-Write Merge: Striped CoW for O(delta) epoch transitions
• AVX-512 Scan Kernel: 64-byte SIMD with 8-row unrolling

License

MIT