Andrea Della Corte ~ posts about books coaching newsletters rss

System Design

On September 3, 2018 in interviewing • 2 minutes read

An algorithm for solving system design interviews:

1. Requirements
   • Extract use cases
   • Understand: Users x Performances x ACID x CAP
2. Math
   • Data in (Concurrency) X Data out
   • Latency x Throughput
3. Main bottlenecks
   • Availability
   • Performance (Response time, Scalability, CPU / IO / Network bound applications)
   • Confidentiality (Encryption)
4. High level design
   • Evaluate use cases: what is the flow for each of them?
   • This usually results in, at a minimum:
   • Application layer & responsibilities
   • Database layer & responsibilities
5. Detailed design
   • Start small & grow big
   • Profiling: Estimate & test load for every use case and identify bottle necks
     • Requests per second resulting in:
       • Write per second (Where? What?)
       • Read per second (Where? What?)
       • Handy reference: 2.5 million seconds per month
   • Horizontal X Vertical Scaling - DNS Server: Resource allocation
   • Resolves the text URL for a particular web resource to the TCP-IP address of the system or service: Must be quick
   • Directs to either:
     • Proxies / Firewalls
     • CDNs: Geographically distributed for static assets: templates , themes , images, etc.
     • Cloud Backend: Web & App functionalities - Load Balancing: Horizontal scaling & Redundancy
   • Software X Hardware
   • Zero Uptime & Increased Performances:
   • Responsibilities:
     • Health checks
     • Load distribution algorithms
   • Challenge: Session Management
     • Sticky sessions X External storage - Web application layer: Serves dynamic content & renders HTML
   • Multiple instances serve independent requests
   • Off-Line Processing: Reduces latency and/or handles batch processing
     • Message Queues: Queue work & process in parallel
     • Scheduling System Tasks: Perform recurring tasks offline
     • Specialized infrastructure: Map-Reduce for big data - Content Performances: Improves use of resources
   • Caching
     • Where?
       • Which layer? Application X Dedicated X Database
       • In memory for vertical scaling?
     • Writethrough cache: Write to cache and then continously push to DB
     • Challenge: Concurrency & Cache Invalidation at App Layer - Manageability: Platform & Management Layer
   • Separates the DB and Web application: Scale the pieces independently.
     • Independent API: Re-use layers for different purposes
   • Includes:
     • Automation & Cost Improvement:
       • Just-in-time Infrastructure
       • Reduces human interaction & errors
     • Monitoring and Alerts
     • Log files
   • Development practices:
     • Source control
     • Multi-step deployment - Database Layer
   • Type: Relational X Graph X Key-Value stores
   • Availability: Master and Standby
   • Performances:
     • Master and Read Replicas
     • Horizontal Scaling of Data Storage: Sharding for storing data on separate databases
     • Per Table Indexes: Avoid searches in your data
   • Data Loss: Backups
6. Security
   • Think of Confidentiality / Integrity / Availability
   • Prevention: Lock out attackers
   • Shared vs Dedicated Instances
   • Access Control & Authentication:
     • Active Directory
     • Two factor authentication
   • Firewalls (Security Groups) between layers of architecture
   • Data & Traffic Encryption - Detection: Find anomalous behaviour
   • Baselining
   • DDoS mitigation - Reaction: Admins & System take appropriate measure to stop attack
   • Alarms
   • Honeypots
7. Low Level
   • Database Schema
   • RESTful API