🖥️ bostrom infrastructure servers

• Servers

  • Back to bostrom infrastructure

• Overview

  • The Bostrom infrastructure runs on 7 dedicated and cloud servers hosted at Hetzner data centers in Germany and Finland.

• Server Fleet

• Deimos — Archive Node

  • Role: Full blockchain history + cyberindex GraphQL API
  • Type: Dedicated server with GPU
  • Location: Falkenstein, Germany
  • Hardware:
    • CPU: Intel Xeon (8 cores)
    • RAM: 64 GB
    • GPU: NVIDIA GTX 1080 8GB (required for consensus)
    • Storage: 6.8 TB ZFS (RAID + NVMe)
  • Services:
    • bostrom_arch — Archive node container (full history from genesis)
    • cyberindex — Blockchain indexer
    • hasura — GraphQL API engine
    • postgres — Index database
  • Endpoints powered:
    • index.bostrom.cybernode.ai — GraphQL API

• Jupiter — RPC Node + IBC Relayer

  • Role: Public RPC/LCD/gRPC endpoints + IBC packet relay
  • Type: Dedicated server with GPU
  • Location: Falkenstein, Germany
  • Hardware:
    • CPU: Intel Xeon (8 cores)
    • RAM: 64 GB
    • GPU: NVIDIA GTX 1080 8GB (required for consensus)
    • Storage: 6.9 TB (ZFS + NVMe)
  • Services:
    • bostrom_pruned — Pruned RPC node (recent state only)
    • space-pussy-arch — Space Pussy testnet archive
    • Hermes IBC relayer (systemd service)
  • Endpoints powered:
    • rpc.bostrom.cybernode.ai — Tendermint RPC
    • lcd.bostrom.cybernode.ai — Cosmos REST API
    • grpc.bostrom.cybernode.ai — gRPC-Web proxy

• IO — IPFS Storage

  • Role: Content-addressed storage for knowledge graph
  • Type: Dedicated server
  • Location: Falkenstein, Germany
  • Hardware:
    • CPU: Intel Xeon
    • RAM: 64 GB
    • Storage: 3.5 TB HDD RAID1
  • Services:
    • IPFS daemon (Kubo 0.22)
    • IPFS cluster (replication)
    • libp2p swarm bootstrap
  • Endpoints powered:
    • gateway.ipfs.cybernode.ai — Public IPFS gateway
    • swarm.io.cybernode.ai — libp2p bootstrap

• Cyberproxy — Reverse Proxy + Frontend

  • Role: SSL termination, load balancing, static hosting
  • Type: Cloud VPS (CX31)
  • Location: Nuremberg, Germany
  • Hardware:
    • vCPU: 2 cores
    • RAM: 8 GB
    • Storage: 80 GB SSD
  • Services:
    • Nginx reverse proxy
    • cyb.ai static frontend
    • Webhook receiver for CI/CD
  • Endpoints powered:
    • cyb.ai — Web interface
    • *.bostrom.cybernode.ai — API proxy

• Mimas — Monitoring

  • Role: Metrics collection, visualization, alerting
  • Type: Cloud VPS (CX41)
  • Location: Falkenstein, Germany
  • Hardware:
    • vCPU: 4 cores
    • RAM: 16 GB
    • Storage: 160 GB SSD
  • Services:
    • Grafana 12.3 — Dashboards and alerting
    • Prometheus 2.54 — Metrics collection
    • Blackbox exporter — HTTP/SSL probing
  • Endpoints powered:
    • cybernode.ai — Public status dashboard
    • cybernode.ai/grafana/ — Grafana dashboards

• Port — API Services

  • Role: Market data and DEX APIs
  • Type: Cloud VPS (CX21)
  • Location: Falkenstein, Germany
  • Hardware:
    • vCPU: 2 cores
    • RAM: 4 GB
    • Storage: 40 GB SSD
  • Services:
    • market-data API — Token supply endpoints
    • warp-dex API — DEX trading data (when available)
  • Endpoints powered:
    • market-data.cybernode.ai — Supply/price data

• Helia-Relay — libp2p Circuit Relay

  • Role: Browser-to-browser IPFS connectivity
  • Type: Cloud VPS (CX32)
  • Location: Helsinki, Finland
  • Hardware:
    • vCPU: 4 cores
    • RAM: 8 GB
    • Storage: 80 GB SSD
  • Services:
    • libp2p circuit relay
    • Caddy (auto-renewing SSL)
  • Purpose:
    • Enables browsers to connect to IPFS network
    • Relay for NAT-traversal

• Hardware Summary

  • Server	Type	RAM	Storage	GPU
    Deimos	Dedicated	64 GB	6.8 TB	GTX 1080
    Jupiter	Dedicated	64 GB	6.9 TB	GTX 1080
    IO	Dedicated	64 GB	3.5 TB	—
    Cyberproxy	Cloud	8 GB	80 GB	—
    Mimas	Cloud	16 GB	160 GB	—
    Port	Cloud	4 GB	40 GB	—
    Helia-Relay	Cloud	8 GB	80 GB	—

• GPU Requirement

  • Bostrom uses GPU-accelerated PageRank for the knowledge graph ranking algorithm.
  • Minimum: NVIDIA GTX 1080 (8GB VRAM)
  • CUDA: 11.4+
  • Both Deimos and Jupiter run GPU nodes for consensus participation.
  • #+BEGIN_NOTE Without GPU, a node cannot participate in consensus or validate blocks. Read-only nodes (for querying) can run without GPU. #+END_NOTE

• Storage Architecture

  • ZFS

    • Deimos and Jupiter use ZFS for data integrity and snapshots
    • Automated snapshots every 12 hours with 2-day retention
    • Manual snapshots before major upgrades

  • Data Volumes

    • Data	Size	Location
      Bostrom archive	~4.4 TB	Deimos
      Bostrom pruned	~500 GB	Jupiter
      Cyberindex DB	~200 GB	Deimos
      IPFS blocks	~1.7 TB	IO

• Software Versions

  • Component	Version
    go-cyber	v7.0.1
    Cosmos SDK	v0.47.16
    CometBFT	v0.37.18
    CosmWasm	v1.5.2
    IPFS (Kubo)	0.22.0
    Hermes	1.13.2
    Grafana	12.3.1
    Prometheus	2.54.1

• Redundancy & Backup

  • Component	Strategy
    Blockchain data	ZFS snapshots + archive node
    IPFS content	Cluster replication
    Configuration	Git-managed
    Cloud servers	Hetzner snapshots

• Related

  • bostrom architecture
  • bostrom monitoring
  • API endpoints