Communication Patterns in Distributed Systems

Back to Distributed Systems
Arjit Sharma
02 Mins

Distributed systems often need to exchange data between services running on different machines, platforms, or programming languages. To make this possible, data must be converted into a transferable format.

Data Serialization & Deserialization

  • Serialization is the process of converting an in-memory data structure into a format suitable for transmission or storage.

  • Deserialization is the reverse process of reconstructing the original data structure from the serialized format.

Common Serialization Formats

JSON (JavaScript Object Notation)

  • Human-readable and easy to debug.
  • Widely used in REST APIs and web applications.
  • Language-independent and supported by virtually every programming language.
  • Larger payload size compared to binary formats, resulting in higher network overhead.

Protocol Buffers (Protobuf)

  • Compact binary serialization format developed by Google.
  • Requires a predefined schema (.proto files).
  • Faster serialization and deserialization than JSON.
  • Reduces bandwidth usage and is commonly used with gRPC.

Choosing the right serialization format involves balancing readability, performance, and compatibility requirements.

Data Transfer Models

Data transfer models define how services communicate and exchange information in a distributed system.

Synchronous Communication (Request–Response)

It is HTTP-based(REST, gRPC) where client sends a request and waits for response. It can cause latency issues.

Common technologies include:

  • HTTP/REST APIs
  • gRPC
  • GraphQL

Example: A frontend application calling a user-service API to fetch profile information.

Asynchronous (Message Driven Model)

It uses message queues (RabbitMQ, Kafka) or event logs where Producer sends a message without waiting for a response and consumer keeps processing messages at its own pace.

Common technologies include:

  • RabbitMQ
  • Apache Kafka
  • Amazon SQS
  • NATS

Example: An e-commerce service publishing an OrderCreated event that is consumed by inventory, payment, and notification services.

Communication Models

Communication models describe how messages are routed between participants in a distributed system.

Point-to-Point (P2P)

In a point-to-point model, a message is sent directly from one sender to one receiver.

Characteristics

  • One producer → one consumer.
  • Simple routing and processing.
  • Commonly used in task queues and job-processing systems.

Example: A payment processing worker consuming jobs from a queue.

Publish-Subscribe (Pub-Sub)

In the publish-subscribe model, publishers send messages to a topic, while subscribers receive messages for topics they are interested in.

Characteristics

  • Publishers and subscribers are decoupled.
  • A single message can be delivered to multiple consumers.
  • Supports event-driven architectures.

Example: A user registration event triggering email, analytics, and auditing services simultaneously.

Next →