Architecture Logicielle et Monolithes

Introduction to Scaling on AWS

• The presentation is about scaling up to 10 million users on Amazon Web Services (AWS).
• The presenter is Giulio Lennazzo, a Solution Architect at Amazon Web Services.

AWS Global Infrastructure

• AWS has 34 regions and 108 availability zones (AZs).
• It has over 600+ points of presence.
• Private backbone network connects these regions.
• Regions are used to define data location and comply with legal risks (e.g., GDPR) enabling disaster recovery.

Availability Zones (AZs)

• AZs are clusters of data centers.
• Each region has at least 3 AZs.
• The AZs provide protection against technical failures.

Points of Presence

• Used to host content delivery services (e.g. Amazon CloudFront).
• These are not directly usable to users.

AWS Local Zones

• Provide region extension, located in major urban and industrial centers.

AWS Wavelength

• 5G infrastructure in stadiums or other high-traffic areas enables fast data access
• Features data analytics, processing and streaming.

AWS Outposts

• Extends AWS infrastructure to on-premises data centers.
• Enables ultra-low latency application, local data processing, and local data storage.

Breadth and Depth of AWS Services

• Over 200 fully featured services are available.
• Services span across several categories (Analytics, Business Applications, Blockchain, Security, Identity...)
• AWS provides comprehensive tools, such as Database, Development, Compute and Media services.

Considerations for Scale

• Decisions are often reversible, enabling flexibility.
• Essential tools for monitoring and managing scale include: Build, Measure, and Learn.
• Important to identify and avoid undifferentiated heavy lifting, choose appropriate solutions (serverless versus managed versus run yourself), and define control vs. responsibility well.

User Scaling Considerations

• Single instance deployment has no failover, redundancy or independent component scaling.
• Deployment options for different user counts (1, >1, >1000) exist.
• Consider managed relational and NoSQL database options to adjust to user scaling needs.
• Amazon RDS offers a managed relational database solution, capable of handling daily maintenance, security patches, and software updates for various database engines (MySQL/MariaDB, PostgreSQL, Oracle, SQLServer, Amazon Aurora).
• Amazon Aurora stands out with MySQL or PostgreSQL compatibility, automatic storage scaling, numerous read replicas and continuous backups.
• Serverless v2 database for adaptable workloads. Deploy on-demand, scaling automatically when needed.

Database Options and Strategies

• Decide between self-managed (Amazon EC2) or fully-managed database options (Amazon RDS, Amazon Aurora, Amazon DynamoDB, etc.).
• NoSQL solutions are favored for high throughput, latency-sensitive, or large volume applications.
• Differentiate cases where NoSQL might be necessary; latency-sensitive applications with required sub-milliseconds response time; terabytes of data volumes and rapid data ingestion; and the use of highly non-relational data like graphs.

AWS Cognito

• A managed user directory service, handling authorization and identity.
• Provides hosted user interfaces for authentication.
• Enables integration with various providers. (e.g., Facebook, Google)

Horizontal Scaling

• Scalability by increasing the number of instances for application load balancing.
• Using load balancers (Application Load Balancer and Network Load Balancer) to distribute the workload evenly across multiple instances in several availability zones (multi-AZ).

Data Consistency

• Crucial to ensure data consistency when scaling (especially with read/write ratios).
• Implementing data replication helps manage data consistency issues. Using multiple regions can address these issues.

Typical read/write ratio

• The ratio of read to write operations, typically with a majority of reads.

Amazon S3

• Object-based storage service, enabling high durability, scaling and storage sizes up to 5TB.

Amazon CloudFront

• Content Delivery Network (CDN) for faster content delivery and handling static asset storage. (e.g., Images, videos, static content).

Caching

• Caching strategies can optimize response times and reduce server load. Caching both inside and outside infrastructure for faster query response times.

Amazon ElastiCache

• Managed caching service (Memcached or Redis) enables quick responses (often single-digit milliseconds).

Amazon DynamoDB

• Managed NoSQL database, offering fast, predictable performance, fault tolerance, and global multi-region tables.

Application Modernization

• Utilize containers and serverless to modernize web applications
• Adopt serverless features (including APIs, queuing, transcoding, search capabilities from databases, monitoring, logging and machine learning) to optimize application performance.

Loose Coupling

• This approach facilitates scalability and resilience by decoupling various components using Amazon SQS and Amazon SNS.

Event-Driven Compute (AWS Lambda)

• Trigger functions automatically based on events within infrastructure.

Microservices Architecture

• Breakdown of large monolithic applications into independent services, enhancing scalability and allowing for independent scaling of individual components.

AWS X-Ray

• Application performance monitoring service for diagnosis of bottlenecks and issues.

Key Recommendations

• Multi-AZ (multi-availability zone) is key for infrastructure resilience and redundancy
• Leverage self-scaling services for optimization (Load Balancer, S3, Lambda, SNS, SQS, CloudFront, etc.)
• Start by selecting appropriate database designs (e.g., relational versus NoSQL).
• Favor automation tools and software development best practices, including the AWS Command Line Interface (CLI) and AWS Cloud Development Kit (CDK) for infrastructure automation.
• Consider caching when necessary for performance improvements.
• Use good metrics and logging for performance monitoring.
• Employ SOA (service-oriented architecture), splitting tiered architectures into independent services, while selecting correct runtimes like Lambda or EC2 based on use cases.

Important Considerations for > 1 Million Users

• Fine-tune existing solutions, especially performance-critical ones.
• Database performance might become an issue.
• Evaluate multi-region deployments for service availability.

Purpose-built Databases

• Use database solutions where applicable (graph, ledger, geo tracking or key-value).
• Leverage database solutions (e.g., NoSQL) when appropriate.

Database Issues & Solutions

• Database issues (potential solutions) could arise and include federating, sharding or employing purpose-built databases or cross-region replication (multi-region).

AWS Amplify

• AWS Amplify enables building apps quickly through UI, integration of backend and front end.

Student Resources

• The presentation contains links to free resources, including AWS Amplify and SageMaker for developers and others interested in learning data science.

Conclusion

• The presentation provides guidelines for scaling up to 10 million users using AWS services.