Skip to content

Golden Rules for System Design Interviews

From: (Dinesh Varyani)[]

Sharing with you 30 golden rules to answer in System Design Interviews. The rules are as follows -

  1. If we are dealing with a read-heavy system, it's good to consider using a Cache.

  2. If we need low latency in the system, it's good to consider using a Cache & CDN.

  3. If we are dealing with a write-heavy system, it's good to use a Message Queue for async processing

  4. If we need a system to be an ACID complaint, we should go for RDBMS or SQL Database

  5. If data is unstructured & doesn't require ACID properties, we should go for No-SQL Database

  6. If the system has complex data in the form of videos, images, files etc, we should go for Blob/Object storage

  7. If the system requires complex/heavy pre-computation like a news feed, we should use a Message Queue & Cache

  8. If the system requires searching data in high volume, we should consider using a search index, tries or a search engine like Elasticsearch

  9. If the system requires to Scale SQL Database, we should consider using Database Sharding

  10. If the system requires High Availability, Performance, & Throughput, we should consider using a Load Balancer

  11. If the system requires faster data delivery globally, reliability, high availability, & performance, we should consider using a CDN

  12. If the system has data with nodes, edges, and relationships like friend lists, & road connections, we should consider using a Graph Database

  13. If the system needs scaling of various components like servers, databases, etc, we should consider using Horizontal Scaling

  14. If the system requires high-performing database queries, we should use Database Indexes

  15. If the system requires bulk job processing, we should consider using Batch Processing & Message Queues

  16. If the system requires reducing server load and preventing DOS attacks, we should use a Rate Limiter

  17. If the system has microservices, we should consider using an API Gateway (Authentication, SSL Termination, Routing etc)

  18. If the system has a single point of failure, we should implement Redundancy in that component

  19. If the system needs to be fault-tolerant, & durable, we should implement Data Replication (creating multiple copies of data on different servers)

  20. If the system needs user-to-user communication (bi-directional) in a fast way, we should use Websockets

  21. If the system needs the ability to detect failures in a distributed system, we should implement a Heartbeat

  22. If the system needs to ensure data integrity, we should use Checksum Algorithm

  23. If the system needs to scale servers with add/removal of nodes efficiently, with no hotspots, we should implement Consistent Hashing

  24. If the system needs to transfer data between various servers in a decentralized way, we should go for Gossip Protocol

  25. If the system needs anything to deal with a location like maps, nearby resources, we should consider using Quadtree, Geohash etc

  26. Avoid using any specific technology names such as - Kafka, S3, or EC2. Try to use more generic names like message queues, object storage etc

  27. If High Availability is required in the system, it's better to mention that system cannot have strong consistency. Eventual Consistency is possible

  28. If asked how domain name query in the browser works and resolves IP addresses. Try to sketch or mention about DNS (Domain Name System)

  29. If asked how to limit the huge amount of data for a network request like youtube search, trending videos etc. One way is to implement Pagination which limits response data.

  30. If asked which policy you would use to evict a Cache. The preferred/asked Cache eviction policy is LRU (Least Recently Used) Cache. Prepare around its Data Structure and Implementation.