Win-win: Maximize Amazon EC2 savings while improving performance (CMP214)

Key Takeaways

Savings and Efficiency

1. Savings has two aspects:

   • Minimizing unnecessary costs by turning off unused instances
   • Ensuring optimal resource utilization to balance performance and cost

2. The promise of the cloud is elasticity, where resources can scale up and down based on demand. This requires the right autoscaling policies and techniques.

Autoscaling and Scaling Strategies

1. Dynamic scaling: Scaling based on actual demand metrics like CPU or memory utilization.

   • Step scaling: Scaling based on thresholds.
   • Target tracking: Scaling to maintain a target metric.

2. Scheduled scaling: Scaling based on predictable events or patterns.

3. Predictive scaling: Scaling based on machine learning models and past behavior.

Instance Optimization

1. AWS has expanded from a single instance type (M1) to over 800 different instance types.
2. Nitro is a hardware-based virtualization platform that enables efficient instance provisioning.
3. Graviton is AWS's Arm-based chip that offers 20-40% better price-performance compared to x86 instances.

Compute Purchase Options

1. On-Demand: No commitments, pay as you go.
2. Savings Plans: Up to 72% savings with 1-3 year commitments.
3. Spot: Up to 90% discount, but instances can be interrupted.

New Bank's Journey

1. New Bank, a large digital bank in Latin America, has adopted a multi-pronged efficiency strategy:

   • Leveraging Spot Instances
   • Migrating to Graviton
   • Diversifying instance types
   • Applying canonical approaches for consistent changes

2. Key lessons learned:

   • Understand service startup times and interruption rates for Spot
   • Determine the optimal Spot discount threshold
   • Diversify instance types to maximize Spot availability
   • Consider the green advantages of Graviton

3. Next steps for New Bank:

   • Developing more nuanced compute purchase strategies as the business expands.