System Design

How to answer in an interview

System design is the process of defining the architecture, components, and data flow of a software system to meet specified functional and non-functional requirements like scalability, reliability, and performance. It's important because good design decisions early on prevent costly rework and ensure a system can handle growth in users and data.

Explain like I am 5

System design is like planning a city's layout before building it — deciding where roads, water pipes, and buildings go so everything works smoothly even as more people move in.

How to answer in an interview

Vertical scaling means adding more resources, like CPU or RAM, to a single existing server, which is simple but has physical limits and creates a single point of failure. Horizontal scaling means adding more machines to distribute the load, which offers better fault tolerance and near-limitless growth but requires more complex coordination like load balancing.

Explain like I am 5

Vertical scaling is like making one worker stronger by giving them more muscles. Horizontal scaling is like hiring more workers to share the job instead.

How to answer in an interview

A CDN is a network of geographically distributed servers that cache and deliver content, like images, videos, and static files, from a location closer to the end user, reducing latency and offloading traffic from the origin server. It's widely used to speed up website loading times and handle traffic spikes.

Explain like I am 5

A CDN is like having copies of a popular book stored in libraries all over the world, so you can borrow it from the nearest library instead of waiting for it to ship from one far-away place.

How to answer in an interview

Latency is the time it takes for a single request to complete, usually measured in milliseconds, while throughput is the number of requests a system can process within a given time period, usually measured in requests per second. A system can have low latency but low throughput, or high throughput but high latency, so both metrics matter for performance evaluation.

Explain like I am 5

Latency is how long it takes one car to drive through a tunnel. Throughput is how many cars can get through the tunnel every minute — they're related but not the same thing.

How to answer in an interview

A load balancer distributes incoming network traffic across multiple servers to prevent any single server from being overwhelmed, improving availability and reliability. Common algorithms include round robin, least connections, and IP hashing, and load balancers often perform health checks to route traffic only to healthy servers.

Explain like I am 5

A load balancer is like a host at a restaurant seating guests evenly across all the waiters, so no single waiter gets overloaded while others stand idle.

How to answer in an interview

Caching stores frequently accessed data in fast storage, like memory, to reduce load on slower backend systems and improve response times. Cache invalidation strategies include time-based expiration (TTL), write-through caching where the cache updates alongside the database, and cache-aside where the application explicitly loads and invalidates cache entries.

Explain like I am 5

Caching is like keeping your most-used snacks on the kitchen counter instead of walking to the store every time. Cache invalidation is knowing when your stashed snacks have gone stale and need replacing.

How to answer in an interview

SQL databases suit systems needing strong consistency, complex relationships, and transactional guarantees, like banking systems. NoSQL databases suit systems prioritizing horizontal scalability, flexible schemas, and high write throughput, like social media feeds or IoT data, though often at the cost of some consistency guarantees.

Explain like I am 5

Choosing SQL vs NoSQL in system design is like choosing between a strict, organized ledger versus a flexible sticky-note board — pick based on whether you need strict accuracy or fast, flexible growth.

How to answer in an interview

A message queue enables asynchronous communication between services by allowing a producer to send messages that a consumer processes independently, decoupling the two components and improving fault tolerance. It's used to smooth out traffic spikes, enable retrying failed operations, and allow services to scale independently, with tools like RabbitMQ and Kafka being common implementations.

Explain like I am 5

A message queue is like a mailbox between two people who don't need to be available at the same time — one drops off a letter whenever, and the other picks it up and reads it whenever they're ready.

How to answer in an interview

A monolithic architecture bundles all application functionality into a single deployable unit, which is simpler to develop and deploy initially but harder to scale and maintain as it grows. Microservices architecture breaks the application into small, independently deployable services communicating over APIs, improving scalability and team autonomy but adding operational complexity like service discovery and distributed monitoring.

Explain like I am 5

A monolith is like one giant machine where every part is bolted together. Microservices are like separate smaller machines that talk to each other, so you can fix or upgrade one without shutting down the rest.

How to answer in an interview

Rate limiting restricts the number of requests a client can make within a given time window to protect a system from abuse or overload. Common algorithms include the token bucket, where tokens refill at a fixed rate and each request consumes one; the leaky bucket, which processes requests at a constant rate; and sliding window counters, which track requests within a moving time frame for more precise limiting.

Explain like I am 5

Rate limiting is like only allowing a certain number of people through a door every minute, so the room inside doesn't get too crowded too fast.

How to answer in an interview

A reverse proxy sits in front of backend servers and forwards client requests to them, often handling tasks like load balancing, SSL termination, caching, and hiding the internal server architecture from clients. Tools like Nginx and HAProxy are commonly used as reverse proxies.

Explain like I am 5

A reverse proxy is like a receptionist who takes your request and passes it to the right person behind the scenes, without you ever needing to know exactly who handled it.

How to answer in an interview

Database replication involves copying data from a primary database to one or more replica databases, often used to improve read scalability by routing read queries to replicas, and to provide high availability by promoting a replica to primary if the original fails. Replication can be synchronous, ensuring replicas are always up to date at the cost of latency, or asynchronous, which is faster but risks slight data lag.

Explain like I am 5

Replication is like keeping several backup copies of an important notebook in different rooms, so if one copy gets lost or someone's busy reading it, you can still check another copy.

How to answer in an interview

An API gateway acts as a single entry point for client requests in a microservices architecture, handling cross-cutting concerns like routing to the appropriate service, authentication, rate limiting, and request/response transformation, so individual services don't need to duplicate this logic.

Explain like I am 5

An API gateway is like the front desk of a big office building — visitors check in once at the front desk, which then directs them to the right department instead of visitors wandering around looking for it themselves.

How to answer in an interview

To eliminate a single point of failure, you introduce redundancy by running multiple instances of critical components across different servers or availability zones, combined with automatic failover mechanisms and health checks so traffic reroutes if one instance goes down. Load balancers, database replicas, and multi-region deployments are common techniques.

Explain like I am 5

Avoiding a single point of failure is like having a spare tire in your car — if one part breaks, there's already a backup ready to take over instead of everything grinding to a halt.

How to answer in an interview

Sharding is a database partitioning technique that splits a large dataset across multiple servers, or shards, based on a shard key, allowing the system to scale horizontally beyond what a single database server could handle. Challenges include choosing a good shard key to avoid hotspots and handling cross-shard queries or joins.

Explain like I am 5

Sharding is like splitting one giant filing cabinet into several smaller cabinets in different rooms, based on some rule like last name, so no single cabinet gets overloaded.

How to answer in an interview

The CAP theorem states that a distributed system can only guarantee two out of three properties simultaneously: Consistency, meaning every read gets the most recent write; Availability, meaning every request gets a response; and Partition tolerance, meaning the system continues to function despite network failures between nodes. Since network partitions are inevitable in distributed systems, designers must choose between prioritizing consistency (CP) or availability (AP).

Explain like I am 5

CAP theorem says that if a group of computers can't talk to each other for a moment, you have to pick: either everyone gives the exact same answer but some might not answer at all, or everyone answers quickly but might give slightly different answers.

How to answer in an interview

I'd generate a unique short code for each long URL, either using a hash function truncated to a fixed length with collision checks, or an auto-incrementing ID encoded in base62. I'd store the mapping in a database with the short code as a key for fast lookups, add a cache layer like Redis for frequently accessed URLs, and consider read replicas since reads vastly outnumber writes in this system.

Explain like I am 5

A URL shortener is like giving a long address a short nickname, then keeping a giant phonebook that remembers which nickname points back to which real address, so whenever someone uses the nickname, we look it up quickly.

How to answer in an interview

Eventual consistency is a consistency model used in distributed systems where updates propagate to all replicas over time, meaning reads immediately after a write might return stale data, but given enough time without new updates, all replicas will converge to the same value. It's a trade-off often chosen in AP systems to favor availability and partition tolerance over strict consistency.

Explain like I am 5

Eventual consistency is like sending the same update to several friends by text — for a moment some friends might not have seen it yet, but eventually everyone will know the same thing.

How to answer in an interview

Consistent hashing maps both data and servers onto a virtual ring using a hash function, so that when a server is added or removed, only a small fraction of keys need to be remapped, rather than rehashing the entire dataset. It's widely used in distributed caching systems and databases to minimize disruption during scaling events.

Explain like I am 5

Consistent hashing is like arranging people and lockers in a big circle — if one locker disappears, only the few people right next to it need a new locker, instead of everyone having to move.

How to answer in an interview

I'd implement a token bucket algorithm where each client has a bucket that refills tokens at a fixed rate, and each request consumes a token, rejecting requests when the bucket is empty. For a distributed system with multiple servers, I'd store the token counts in a fast shared store like Redis to keep limits consistent across all instances, using atomic operations to avoid race conditions.

Explain like I am 5

A rate limiter is like giving someone a set number of tickets per hour — once their tickets run out, they have to wait until new tickets are handed out before they can enter again.