Write-through and write-back caching in HLD - System Design Exercise

Design: Caching System with Write-through and Write-back Strategies

Design focuses on caching strategies and data flow between cache and database. Does not cover specific database or cache technology selection or network infrastructure.

Functional Requirements

FR1: Support caching of data to improve read and write performance

FR2: Implement write-through caching where writes update both cache and database synchronously

FR3: Implement write-back caching where writes update cache first and database asynchronously

FR4: Ensure data consistency between cache and database

FR5: Handle cache misses by fetching data from the database

FR6: Support configurable cache eviction policies

FR7: Provide metrics on cache hit rate and write latency

Non-Functional Requirements

NFR1: System should handle 10,000 concurrent read/write requests

NFR2: Read latency p99 should be under 50ms

NFR3: Write latency p99 should be under 100ms for write-through and under 20ms for write-back

NFR4: Availability target of 99.9% uptime

NFR5: Data loss risk must be minimized especially for write-back caching

Think Before You Design

Questions to Ask

❓ Question 1

❓ Question 2

❓ Question 3

❓ Question 4

❓ Question 5

Key Components

Cache layer (in-memory store like Redis or Memcached)

Primary database (relational or NoSQL)

Write queue or buffer for asynchronous writes in write-back

Cache eviction mechanism

Monitoring and metrics system

Design Patterns

Write-through caching pattern

Write-back (write-behind) caching pattern

Cache-aside pattern for cache misses

Asynchronous processing with queues

Data consistency and synchronization techniques

Reference Architecture

Client
  |
  v
Cache Layer (supports write-through and write-back)
  |        |
  |        v
  |    Write Queue (for write-back)
  |        |
  v        v
Database <---- Monitoring & Metrics

Components

Cache Layer

In-memory store (e.g., Redis, Memcached)

Serve fast read requests and handle writes according to caching strategy

Primary Database

Relational or NoSQL database

Persistent data storage

Write Queue

Message queue or buffer (e.g., Kafka, RabbitMQ)

Buffer writes asynchronously in write-back caching

Monitoring & Metrics

Prometheus, Grafana or similar

Track cache hit rates, write latencies, and system health

Request Flow

1. Client sends read or write request to Cache Layer.

2. For reads: Cache Layer checks cache; if hit, returns data immediately.

3. If cache miss, Cache Layer fetches data from Database, updates cache, then returns data.

4. For writes in write-through mode: Cache Layer writes data to cache and synchronously to Database before confirming success to client.

5. For writes in write-back mode: Cache Layer writes data to cache and enqueues write operation to Write Queue, then immediately confirms success to client.

6. Write Queue processes queued writes asynchronously and updates Database.

7. Monitoring system collects metrics on cache hits, misses, write latencies, and errors.

Database Schema

Entities: DataItem(id PK, value, last_updated) Relationships: None specific to caching; schema supports data storage and timestamp for synchronization.

Scaling Discussion

Bottlenecks

Cache capacity limits causing frequent evictions and cache misses

Write Queue becoming a bottleneck under heavy write-back load

Database write throughput limits especially for write-through caching

Data inconsistency risk in write-back due to delayed writes

Monitoring overhead impacting system performance

Solutions

Scale cache horizontally with sharding or clustering to increase capacity

Use multiple partitions or consumers for Write Queue to increase throughput

Optimize database writes with batching and indexing; consider database scaling

Implement retry and failure detection mechanisms for write-back queue to reduce data loss

Use lightweight monitoring agents and sample metrics to reduce overhead

Interview Tips

Time: Spend 10 minutes understanding requirements and clarifying consistency needs, 15 minutes designing architecture and data flow, 10 minutes discussing scaling and trade-offs, 10 minutes answering questions.

Explain difference between write-through and write-back caching clearly

Discuss trade-offs between data consistency, latency, and durability

Describe how cache misses and evictions are handled

Highlight asynchronous processing benefits and risks in write-back

Address scaling challenges and mitigation strategies

Mention monitoring importance for operational visibility