Cost vs Performance Optimization

Autopilot offers two scaling strategies to balance cost savings with performance needs. Choose based on your workload's business criticality and latency requirements.

Strategy Comparison

	Optimize for Performance	Optimize for Cost
Priority	Responsiveness and headroom	Efficiency and savings
Capacity	Keeps spare capacity for spikes	Runs closer to the limit
Best for	User-facing, latency-sensitive, revenue-critical services	Internal tools, batch jobs, latency-tolerant services
Trade-off	Higher cost for stability	Lower cost, less spike headroom

Optimize for Performance

How it works: Keeps higher capacity buffer, rounds up replica calculations, maintains lower utilization targets.

Best for:

User-facing services with strict SLAs (< 100ms p99)
Revenue-critical paths (checkout, payment, authentication)
High-traffic production APIs
Services where latency directly impacts user experience

Example Applications: E-commerce checkout flow, real-time bidding systems, video streaming APIs, authentication/authorization services

Quick Setup

Scaling Strategy: Optimize for Performance
Min Replicas: 5
Max Replicas: 30
Gradual Scaling: 120s
Predictive Scaling: Enabled

Expected Behavior: Maintains 10-20% spare capacity, scales proactively before traffic hits, tolerates brief over-provisioning.

Optimize for Cost

How it works: Targets higher utilization, rounds normally, scales down more aggressively while maintaining reliability.

Best for:

Internal tools and admin dashboards
Background processing and batch jobs
Development/staging environments
Services with relaxed latency requirements (> 500ms acceptable)

Example Applications: Data processing pipelines, report generation services, analytics ingestion, internal APIs and tools

Quick Setup

Scaling Strategy: Optimize for Cost
Min Replicas: 2
Max Replicas: 15
Gradual Scaling: 60s
Predictive Scaling: Disabled

Expected Behavior: Runs closer to capacity limits, scales down quickly during low traffic, accepts brief latency increases during sudden spikes.

Choosing Your Strategy

Use Performance When:

Service has strict SLA requirements
Latency directly impacts revenue or user experience
Traffic spikes are unpredictable
Business prefers over-provisioning to any service degradation

Use Cost When:

Service can tolerate occasional latency spikes
Traffic is predictable or latency-insensitive
Budget optimization is a priority
Service is non-customer-facing

Common Patterns

🎯 Tiered Approach (same application, different services)

Frontend API: Performance (user-facing)
Order Processing: Performance (revenue-critical)
Product Catalog: Cost (cached, less sensitive)
Analytics Pipeline: Cost (background job)

Apply different strategies based on service criticality within the same application.

⏰ Time-Based Approach (same service, different times)

Use Autopilot Scheduler to switch strategies:

Business hours (9 AM - 6 PM): Performance mode
Off-hours (6 PM - 9 AM): Cost mode
Weekends: Cost mode

Automatically adjust optimization based on expected traffic patterns.

Monitoring Your Choice

Watch these metrics after enabling:

⚠️

Performance Mode Warning Signs:

Average CPU consistently below 50%
Long periods with idle replicas
High cloud costs with minimal traffic

Fix: Switch to Cost mode or lower Max Replicas

⚠️

Cost Mode Warning Signs:

P99 latency spikes during traffic increases
Error rates correlate with scaling events
Users reporting slowness

Fix: Switch to Performance mode or enable Predictive Scaling

Quick Decision Guide

💡

Ask yourself: "What happens if this service is slow for 30 seconds?"

Business loses money / Users complain: Use Performance
No one notices / It's acceptable: Use Cost

Example: E-Commerce Site

Service	Strategy	Why
Checkout API	Performance	Direct revenue impact
Product Search	Performance	Core user experience
Product Details	Cost	Cached, less time-sensitive
Order History	Cost	Async, background query
Email Service	Cost	Queued, no real-time requirement

Next Steps

Predictive Scaling: Add forecasting to either strategy for proactive scaling
Autopilot Scheduler: Automatically switch strategies based on time
Timing Controls: Fine-tune scale-down behavior for your strategy

Core Configuration Predictive Scaling