⁠[Proposal] Low-Latency Infrastructure Benchmark: Co-locating Polygon Erigon Full-Node & GPU Routing Engine on Akash VM

[pikapika1774]

[Proposal] Low-Latency Infrastructure Benchmark: Co-locating Polygon Erigon Full-Node & GPU Routing Engine on Akash VM

1. Project Overview

We are a core low-latency infrastructure development initiative focusing on Web3 algorithmic optimization. We have successfully engineered a proprietary, GPU-based multihop swap routing engine capable of executing over 3,500,000 paths within 0.9 milliseconds (900 microseconds) using raw EVM chain data.

The core architecture utilizes a unique I/O bottleneck elimination logic between host memory and GPU CUDA cores. This ensures extreme horizontal scalability: even if the complexity of the search space scales up to 10,000,000+ paths, the calculation execution time scales sub-linearly and remains sub-millisecond, provided sufficient GPU hardware is supplied.

2. The Objective on Akash Network

To unlock the ultimate execution speed needed for high-frequency trading (HFT) and Maximal Extractable Value (MEV) extraction, we aim to co-locate a dense Polygon Erigon Full-Node and our GPU routing engine within the exact same Akash Virtual Machine (VM) instance.

By bypassing external, latent RPC endpoints and streaming raw mempool/block state data directly via local Inter-Process Communication (IPC / Domain Sockets) straight into GPU memory, we will eliminate network lag entirely. We want to benchmark and officially prove to the Web3 community that Akash’s decentralized infrastructure can outperform traditional, high-cost centralized bare-metal setups in the most latency-sensitive vertical in crypto.

3. Infrastructure & Grant Request

Due to the sheer size of the Polygon ledger and the intensive disk I/O required for sub-millisecond data scraping, we are bottlenecked by the upfront infrastructure costs. We are requesting a Seed/Incubator Grant (denominated in AKT) to cover the leasing fees of a high-spec Akash provider for an initial period of 3 months:

• GPU: 1x NVIDIA RTX 4090 (or equivalent consumer/enterprise GPU optimized for high-parallelism CUDA execution)
• Storage: 8TB+ High-Speed NVMe SSD (Strictly required to host the dense Polygon Erigon snapshot and withstand massive continuous disk write/read cycles)
• CPU/RAM: High-frequency multi-core CPU with sufficient RAM (64GB+) to prevent host-to-device bottlenecks.

4. Deliverables & Open-Source Policy

• What we will share with the Akash Community:
  • Comprehensive, granular low-latency benchmark reports.
  • Performance logs explicitly showing the timeline of [Erigon State Ingestion -> GPU Core Processing -> Transaction/Bundle Generation < 0.9ms].
  • The customized SDL (Stack Definition Language) deployment templates, enabling other sophisticated node runners to optimize their high-throughput stacks on Akash.
• Intellectual Property Boundaries:
  To preserve our proprietary algorithmic edge and protect our core IP, the optimized CUDA/C++ routing engine will remain as compiled binaries during this benchmarking phase.

We believe this project will brilliantly showcase the heavy-compute capabilities of Akash Network to institutional-grade developers and quants worldwide.

Thank you for your review and support!