Amazon EC2 Fundamentals for Solutions Architects¶

📖 EC2 User Guide - Complete EC2 documentation and best practices

Instance Types and Selection¶

📖 EC2 Instance Types - Detailed guide to choosing the right instance type for your workload

General Purpose Instances¶

T4g, T3, T3a, T2 - Burstable Performance¶

Use Cases: Web servers, small databases, development environments, code repositories
Performance Model: Baseline performance with ability to burst
CPU Credits: Accumulate credits when below baseline, consume when bursting
Unlimited Mode: T3/T4g can burst beyond credits for additional cost
Sizing Strategy: Monitor CPU credit balance and utilization patterns

📖 Burstable Performance Instances - Understanding CPU credits and burstable performance

T4g vs T3 vs T2: - T4g: ARM-based Graviton2 processors, 20% better price/performance - T3: Latest generation Intel processors, better baseline performance than T2 - T2: Previous generation, lower baseline performance

M6i, M5, M5a, M4 - Balanced Performance¶

Use Cases: Web applications, microservices, enterprise applications, small/mid-size databases
Characteristics: Balanced compute, memory, and networking resources
vCPU Range: 1-96 vCPUs depending on size
Memory Range: 4 GiB - 384 GiB
Network Performance: Up to 25 Gbps

Compute Optimized Instances¶

C6i, C5, C5n, C4 - High Performance Computing¶

Use Cases: CPU-intensive applications, web servers, scientific modeling, batch processing, MMO gaming
Characteristics: High-performance processors, optimized for compute-intensive workloads
Processor Features: Latest Intel processors with high clock speeds
Network Performance: Enhanced networking capabilities

When to Choose Compute Optimized: - CPU utilization consistently above 40% - Compute-bound applications - High-performance web servers - Scientific and financial modeling

Memory Optimized Instances¶

R6i, R5, R5a, R4 - Memory Optimized¶

Use Cases: In-memory databases, real-time analytics, high-performance databases
Memory-to-vCPU Ratio: Optimized for memory-intensive applications
Performance: Fast processors suitable for memory-bound workloads

X1e, X1 - High Memory¶

Use Cases: In-memory databases (SAP HANA), Apache Spark, Presto
Memory Range: Up to 3,904 GiB of memory
SSD Storage: High I/O performance with SSD-backed instance storage

z1d - High Frequency¶

Use Cases: Electronic Design Automation (EDA), relational databases with high per-core licensing costs
Processor: Sustained all-core frequency of up to 4.0 GHz
Storage: NVMe SSD instance storage

Storage Optimized Instances¶

I4i, I3, I3en - NVMe SSD Storage¶

Use Cases: NoSQL databases, in-memory databases, data warehouses, search engines
Storage Type: NVMe SSD instance storage
I/O Performance: Very high random I/O performance
Network Performance: High network performance for data-intensive applications

D3, D2 - Dense HDD Storage¶

Use Cases: Distributed file systems, data processing workloads, log processing
Storage Type: High density HDD storage
Cost: Lower cost per GB compared to SSD options

Accelerated Computing Instances¶

P4, P3 - GPU for Machine Learning¶

Use Cases: Machine learning training, high-performance computing, seismic analysis
GPU: NVIDIA Tesla V100 (P3) or A100 (P4) GPUs
Memory: High bandwidth memory for GPU workloads
Networking: 100 Gbps network performance (P4)

G4 - GPU for Graphics Workloads¶

Use Cases: Game streaming, 3D visualization, video processing
GPU: NVIDIA T4 GPUs
Applications: Graphics workstations, media rendering

Instance Sizing Strategy¶

Right-Sizing Methodology¶

Start Small: Begin with smaller instances and scale up based on actual usage
Monitor Metrics: Use CloudWatch to monitor CPU, memory, network, and disk utilization
Analyze Patterns: Understand peak vs. average utilization patterns
Consider Burstable: Use T-series for variable workloads
Plan for Growth: Consider future capacity needs

Key Metrics for Sizing¶

CPU Utilization: Target 40-60% average utilization
Memory Utilization: Monitor swap usage and available memory
Network Utilization: Consider bandwidth requirements
Disk I/O: Monitor IOPS and throughput requirements

Placement Groups¶

📖 Placement Groups - Strategies for grouping instances to meet workload needs

Cluster Placement Groups¶

Purpose: Low latency, high network performance between instances
Use Cases: HPC applications, tightly coupled workloads
Limitations: Single AZ, limited instance types
Network Performance: 10 Gbps between instances in same cluster placement group

Partition Placement Groups¶

Purpose: Large distributed workloads (Hadoop, Cassandra, Kafka)
Partitions: Up to 7 partitions per AZ
Isolation: Each partition has its own set of racks
Use Cases: Large distributed and replicated workloads

Spread Placement Groups¶

Purpose: Critical instances that should be on separate hardware
Limitation: Maximum 7 instances per AZ per group
Use Cases: Small number of critical instances
Isolation: Each instance on separate underlying hardware

Instance Purchasing Options¶

📖 EC2 Pricing - Understanding different EC2 pricing models

On-Demand Instances¶

Billing: Pay by the hour or second (minimum 60 seconds)
Use Cases: Short-term workloads, unpredictable workloads, development/testing
Benefits: No upfront payment, no long-term commitment
Best For: Applications with unpredictable workloads that cannot be interrupted

Reserved Instances¶

Standard Reserved Instances:
Discount: Up to 75% compared to On-Demand
Flexibility: Can modify AZ, instance size (within same family), networking type
Term: 1 or 3 years
Payment: All Upfront, Partial Upfront, No Upfront
Convertible Reserved Instances:
Discount: Up to 54% compared to On-Demand
Flexibility: Can change instance family, OS, tenancy, payment option
Term: 3 years only
Use Case: Workloads with changing requirements
Scheduled Reserved Instances:
Use Case: Predictable recurring schedules
Commitment: Specific time windows (daily, weekly, monthly)
Discount: Compared to On-Demand for scheduled usage

Spot Instances¶

Pricing: Up to 90% discount compared to On-Demand
Availability: Based on spare EC2 capacity
Interruption: 2-minute warning before termination
Best Practices:
Use Spot Fleet for diversification
Implement checkpointing for long-running jobs
Use multiple instance types and AZs
Set appropriate maximum price

📖 Spot Instances - Maximize savings with EC2 Spot Instances

Ideal Workloads for Spot: - Batch processing jobs - Data analysis and processing - Image and media rendering - Scientific computing - CI/CD and testing

Dedicated Instances¶

Isolation: Run on single-tenant hardware
Use Cases: Compliance requirements, licensing restrictions
Billing: Additional charges apply
Placement: May share hardware with other Dedicated Instances from same account

Dedicated Hosts¶

Control: Physical EC2 server dedicated for your use
Visibility: Socket and core visibility for licensing
Use Cases: Server-bound software licenses, compliance requirements
Management: More control over instance placement

Auto Scaling Strategies¶

📖 Auto Scaling - Maintain application availability with EC2 Auto Scaling

Auto Scaling Groups (ASG)¶

Components¶

Launch Template/Configuration: Defines instance specifications
Auto Scaling Group: Manages collection of instances
Scaling Policies: Rules for scaling actions
Health Checks: EC2 and/or ELB health checks

Scaling Policies¶

Target Tracking Scaling: - Metric: Maintain target value for specific metric - Examples: Average CPU utilization, request count per target - Behavior: ASG automatically adjusts capacity to maintain target - Best Practice: Use for most common scaling scenarios

Step Scaling: - Triggers: CloudWatch alarms trigger scaling actions - Steps: Different scaling amounts based on alarm breach size - Flexibility: More control over scaling behavior - Use Case: When you need fine-tuned control over scaling

Simple Scaling: - Action: Single scaling action per alarm - Cooldown: Wait period before next scaling action - Legacy: Older scaling type, step scaling preferred

Predictive Scaling: - Machine Learning: Uses ML to predict future traffic - Proactive: Scales before traffic increases - Best Practice: Combine with reactive scaling policies

Health Checks¶

EC2 Health Checks: Instance status and system status
ELB Health Checks: Application-level health verification
Custom Health Checks: Via CloudWatch or custom scripts
Grace Period: Time to allow instance to warm up

Best Practices for Auto Scaling¶

Design Principles¶

Stateless Applications: Design applications to be stateless
Health Checks: Implement meaningful health checks
Graceful Shutdown: Handle termination signals properly
Multiple AZs: Distribute instances across multiple AZs
Right-Size: Use appropriate instance types for workload

Monitoring and Optimization¶

CloudWatch Metrics: Monitor ASG and instance metrics
Scaling History: Review scaling activities and effectiveness
Cost Optimization: Balance performance and cost
Testing: Test scaling policies under various load conditions

Security Best Practices¶

📖 EC2 Security - Best practices for securing your EC2 instances

Security Groups¶

Stateful: Return traffic automatically allowed
Default Behavior: Deny all inbound, allow all outbound
Best Practices:
Principle of least privilege
Use descriptive names and descriptions
Regular review and cleanup
Use security group references instead of IP addresses

📖 Security Groups - Control traffic to your instances with security groups

IAM Roles for EC2¶

Instance Profiles: Attach IAM roles to EC2 instances
Benefits: No need to store credentials on instances
Automatic Rotation: Credentials automatically rotated
Best Practice: Always use IAM roles instead of access keys

📖 IAM Roles for EC2 - Grant AWS permissions to applications running on EC2

Key Pairs¶

SSH Access: Use key pairs for Linux instances
RDP Access: Use key pairs or set passwords for Windows
Best Practices:
Use separate key pairs for different environments
Regularly rotate key pairs
Store private keys securely
Use Systems Manager Session Manager when possible

Network Security¶

VPC Placement: Launch instances in VPC with proper subnet design
Network ACLs: Additional layer of network security
Monitoring: Enable VPC Flow Logs for network monitoring
Encryption: Use encryption in transit and at rest

Performance Optimization¶

📖 Enhanced Networking - Enable high network performance for supported instance types

Instance Optimization¶

Enhanced Networking: Enable for supported instance types
SR-IOV: Single Root I/O Virtualization for better network performance
Placement Groups: Use appropriate placement group type
Instance Store: Use for temporary high-performance storage

Monitoring and Tuning¶

CloudWatch Metrics: Monitor detailed instance metrics
Custom Metrics: Send application-specific metrics
Performance Baselines: Establish performance baselines
Load Testing: Regular load testing to validate performance

Storage Performance¶

EBS Optimization: Enable EBS optimization for supported instances
Instance Store: Use for temporary high-IOPS requirements
Network Bandwidth: Consider network bandwidth for storage throughput

Cost Optimization Strategies¶

Instance Selection¶

Right-Sizing: Regular review and adjustment of instance sizes
Instance Families: Choose appropriate instance family for workload
Generation: Use latest generation instances for better price/performance

Purchasing Options¶

Reserved Instances: For predictable workloads
Spot Instances: For fault-tolerant workloads
Savings Plans: For flexible workload patterns

Operational Efficiency¶

Auto Scaling: Automatically adjust capacity based on demand
Scheduling: Stop/start instances for dev/test environments
Resource Tagging: Implement comprehensive tagging strategy
Cost Monitoring: Regular review of cost and usage reports

Common Architecture Patterns¶

Web Application Tier¶

Load Balancer: Application Load Balancer for HTTP/HTTPS traffic
Auto Scaling: Auto Scaling Group across multiple AZs
Instance Type: General purpose (M5) or compute optimized (C5)
Health Checks: Application-level health checks

Application Tier¶

Private Subnets: Deploy in private subnets for security
Auto Scaling: Based on application metrics
Instance Type: Based on application requirements (CPU, memory)
Service Discovery: Use ELB or service mesh for service discovery

Background Processing¶

Spot Instances: Use Spot Instances for cost optimization
Queue-based: Use SQS for decoupling and fault tolerance
Auto Scaling: Scale based on queue depth
Instance Type: Compute or memory optimized based on workload

Key Takeaways¶

Instance Selection: Choose instance types based on workload characteristics (CPU, memory, storage, network)
Purchasing Strategy: Mix of On-Demand, Reserved, and Spot instances for cost optimization
Auto Scaling: Implement Auto Scaling for availability and cost optimization
Security: Use IAM roles, security groups, and VPC for comprehensive security
Performance: Monitor and optimize based on actual usage patterns
Cost Management: Regular review and optimization of instance usage and costs
Architecture: Design for fault tolerance using multiple AZs and Auto Scaling
Monitoring: Comprehensive monitoring for performance and cost optimization