Sizing Guide

Overview

Comprehensive sizing guidance for NeuronDB deployments from development to enterprise scale.

Resource Profiles

Small (Development/Testing)

Use Case: Development, testing, small teams

Workload Capacity:

• Concurrent users: 10-50
• Requests/second: 100-500
• Vector operations: 1K-10K/day

Medium (Production - Small)

Use Case: Small production deployments, <100 users

Workload Capacity:

• Concurrent users: 50-200
• Requests/second: 500-2K
• Vector operations: 10K-100K/day

Large (Production - Medium)

Use Case: Medium production deployments, 100-500 users

Total Resources:

• CPU: ~20 cores
• Memory: ~26Gi
• Storage: 500Gi

Workload Capacity:

• Concurrent users: 200-1000
• Requests/second: 2K-10K
• Vector operations: 100K-1M/day

Enterprise (Production - Large)

Use Case: Large production deployments, 500+ users

Total Resources:

• CPU: ~46 cores
• Memory: ~58Gi
• Storage: 1Ti

Workload Capacity:

• Concurrent users: 1000+
• Requests/second: 10K+
• Vector operations: 1M+/day

Storage Sizing

Base Storage Requirements

• OS and binaries: ~10Gi
• PostgreSQL data: Variable
• WAL files: ~10% of data size
• Logs: ~5Gi per month
• Backups: 2-3x data size (if local)

Growth Projections

Plan storage for 1 year growth:

• Small: 50Gi → 100-150Gi
• Medium: 200Gi → 400-600Gi
• Large: 500Gi → 1-1.5Ti
• Enterprise: 1Ti → 2-3Ti

Storage Class Recommendations

• Development: Standard SSD
• Production: Premium SSD or GP3
• Enterprise: Premium SSD with IOPS optimization

Network Bandwidth

Estimated Requirements

• Small: 100 Mbps
• Medium: 1 Gbps
• Large: 10 Gbps
• Enterprise: 25 Gbps+

Factors Affecting Bandwidth

• Vector embedding size
• Query frequency
• Replication (if enabled)
• Backup operations

CPU Sizing

NeuronDB CPU

• Base: 2 CPU for PostgreSQL
• Vector operations: +1 CPU per 10K vectors/second
• Concurrent queries: +0.5 CPU per 100 concurrent

NeuronAgent CPU

• Base: 500m per replica
• Request handling: +100m per 100 req/s
• Background workers: +200m per worker

NeuronMCP CPU

• Base: 250m per replica
• MCP requests: +50m per 50 req/s

Memory Sizing

NeuronDB Memory

• Base: 4Gi for PostgreSQL
• Shared buffers: 25% of total memory
• Work memory: 256Mi per connection
• Vector cache: 1Gi per 1M vectors

NeuronAgent Memory

• Base: 512Mi per replica
• Request buffer: 100Mi per 100 req/s
• Cache: 200Mi per replica

NeuronMCP Memory

• Base: 256Mi per replica
• Request buffer: 50Mi per 50 req/s

Autoscaling Recommendations

HPA Configuration

neuronagent:
  autoscaling:
    enabled: true
    minReplicas: 2
    maxReplicas: 10
    targetCPUUtilizationPercentage: 70

KEDA Configuration

neuronagent:
  autoscaling:
    keda:
      enabled: true
      minReplicas: 2
      maxReplicas: 20
      triggers:
        http:
          enabled: true
          threshold: "100"
        queueDepth:
          enabled: true
          targetValue: "100"

Node Requirements

Minimum Node Specs

• CPU: 4 cores
• Memory: 8Gi
• Storage: 50Gi
• OS: Linux (Ubuntu 20.04+, RHEL 8+)

Recommended Node Specs (Production)

• CPU: 8+ cores
• Memory: 16Gi+
• Storage: 200Gi+ SSD
• Network: 1 Gbps+