MIZ OKI 3.0 Cloud Run Resource Budget

Version: 1.0
Last Updated: 2025-11-14
Status: Active

Overview

This document defines the CPU and memory allocation budget for all MIZ OKI 3.0 Cloud Run services. Resource allocations are based on service role, computational requirements, and expected load patterns.

Resource Allocation Philosophy

1. Orchestration services (Boss Agent, orchestrators, gateways) receive higher resources to prevent bottlenecks
2. Compute-intensive cells (ML inference, optimization, analytics) receive maximum resources
3. Standard cells receive baseline resources with autoscaling
4. Support services (monitoring, logging) receive moderate resources

Resource Categories

Service-Specific Allocations

Orchestration Layer

SENSE Stage (Cells 1-5)

REASON Stage (Cells 6-9)

DECIDE Stage (Cells 10-13)

ACT Stage (Cells 14-18)

LEARN Stage (Cells 19-25)

Additional Services

Autoscaling Configuration

Orchestration Services

• Min Instances: 2 (high availability)
• Max Instances: 8-10
• Target: 80% CPU utilization
• Concurrency: 150-200 requests

High Compute Cells

• Min Instances: 1
• Max Instances: 20
• Target: 80% CPU utilization
• Concurrency: 80
• CPU Throttling: Disabled

Standard Cells

• Min Instances: 1
• Max Instances: 10
• Target: 80% CPU utilization
• Concurrency: 80-100

Cost Optimization Guidelines

1. Use CPU-only tiers for services without GPU requirements
2. Enable CPU throttling for non-critical services that can tolerate latency
3. Set appropriate min instances: 0 for dev, 1 for staging, 1-2 for production
4. Monitor actual usage: Review Cloud Monitoring metrics weekly
5. Right-size iteratively: Start conservative, scale up based on metrics

Monitoring and Adjustment

Key Metrics to Monitor

1. CPU Utilization - Target: 60-80% average - Alert: >90% for sustained periods

2. Memory Utilization - Target: 60-80% average - Alert: >85% (risk of OOM)

3. Request Latency - Target: p95 <2s for orchestration, <5s for compute - Alert: p95 >5s

4. Instance Count - Monitor actual vs max instances - Adjust max if frequently hitting ceiling

5. Cold Start Frequency - Target: <5% of requests - Consider increasing min instances if higher

Adjustment Process

1. Weekly Review: Check resource utilization dashboard
2. Monthly Planning: Adjust budgets based on usage trends
3. Incident-Driven: Update immediately if bottlenecks occur
4. Quarterly Audit: Comprehensive review with cost optimization

Resource Request Guidelines for New Services

When adding new services, use this decision tree:

Is it an orchestration/coordination service?
├─ YES → Category: Critical Orchestration (2 CPU / 4Gi)
└─ NO
   ├─ Does it perform ML inference or optimization?
   │  ├─ YES → Category: High Compute (4-8 CPU / 8-16Gi)
   │  └─ NO
   │     ├─ Does it process large datasets?
   │     │  ├─ YES → Category: Standard Cell (1-2 CPU / 1-2Gi)
   │     │  └─ NO → Category: Light Service (500m CPU / 512Mi)

Historical Changes

Budget Summary

Total Resource Budget (Production)

With min instances running: - Total CPU: ~50-60 vCPUs - Total Memory: ~60-80 GiB - Estimated Monthly Cost: $500-800 USD

At max instances (worst case): - Total CPU: ~200-250 vCPUs - Total Memory: ~250-300 GiB - Estimated Monthly Cost: $2,000-3,000 USD

Typical Operating Point: - Total CPU: ~80-100 vCPUs - Total Memory: ~100-120 GiB - Estimated Monthly Cost: $800-1,200 USD

Related Documentation

• Service Discovery Configuration
• Deployment Guide
• Scaling Best Practices (to be created)
• Cost Optimization (to be created)

Maintained by: MIZ OKI Platform Team
Review Frequency: Monthly
Next Review: 2025-12-14