Inferensys

Difference

Zenoh vs Cyclone DDS: The Next-Gen Pub/Sub Protocol Challenges the Established DDS Standard

A technical comparison of Zenoh and Cyclone DDS for production-grade robotics middleware. We analyze minimal wire overhead, peer-to-peer scaling, high-throughput reliability, and integration with ROS 2 to help lead robotics engineers choose the right communication fabric for multi-robot fleets and physical AI.
Wide-angle shot of a modern WeWork open floor plan with creative walls covered in AI system architecture diagrams, product team collaborating in standing desk area with industrial lighting.
THE ANALYSIS

Introduction

A data-driven comparison of Zenoh's minimal wire overhead against Cyclone DDS's proven reliability for high-throughput, multi-robot coordination.

Zenoh excels at minimizing wire overhead and scaling peer-to-peer communication across diverse network topologies because it decouples data exchange from rigid discovery protocols. For example, benchmarks show Zenoh achieving over 3.5 million messages per second with a wire overhead of just 5 bytes per data message, making it exceptionally efficient for constrained networks and cloud-to-edge bridging.

Cyclone DDS takes a different approach by fully embracing the Object Management Group's (OMG) Data Distribution Service (DDS) standard, which provides a proven, feature-rich middleware layer. This results in a trade-off: higher wire overhead (typically 30-40 bytes per message) but with built-in, fine-grained Quality of Service (QoS) policies, decentralized discovery, and a decade of battle-tested reliability for coordinating high-throughput, multi-robot fleets over lossy networks.

The key trade-off: If your priority is minimizing bandwidth consumption, bridging non-ROS systems, or scaling across geo-distributed cloud and edge nodes, choose Zenoh. If you prioritize a fully standardized, plug-and-play middleware with mature tooling and deterministic behavior for a tightly coupled fleet of ROS 2 robots, choose Cyclone DDS.

HEAD-TO-HEAD COMPARISON

Feature Comparison Matrix

Direct comparison of key metrics and features for next-gen pub/sub protocols in multi-robot coordination.

MetricZenohCyclone DDS

Wire Overhead (Bytes)

5-6

36+

Peak Throughput (Gbps)

50

~10

Latency P99 (µs)

~30

~50

Native Peer-to-Peer

QoS Reliability

Configurable

Rich (22 policies)

Wire Protocol Compatibility

Pluggable (TCP/UDP/QUIC)

RTPS Standard

ROS 2 RMW Support

Community (rmw_zenoh)

Native (rmw_cyclonedds)

Zenoh vs Cyclone DDS

TL;DR Summary

A high-level breakdown of the core strengths and trade-offs between the next-gen Zenoh protocol and the established Cyclone DDS standard for robotics middleware.

01

Zenoh: Minimal Wire Overhead & Peer-to-Peer Scaling

Wire efficiency: Zenoh's protocol has an extremely low wire overhead (often < 5 bytes for data messages), making it ideal for constrained networks and wireless links. Peer-to-peer architecture: It natively supports peer-to-peer communication without a dedicated discovery service, enabling seamless scaling across dynamic, mesh-like robot swarms. This matters for multi-robot fleets operating over lossy Wi-Fi or 5G, where every byte counts and infrastructure is ad-hoc.

02

Zenoh: Unified Abstraction for Cloud-to-Microcontroller

Protocol versatility: Zenoh provides a single, unified protocol that operates over diverse transports (TCP, UDP, QUIC, Serial) and bridges from deeply embedded microcontrollers to cloud data centers. Key advantage: This eliminates the need for complex protocol bridges (like ROS 2's rosbridge) when integrating micro-ROS devices with cloud analytics. This matters for system architects building a unified data fabric from the sensor edge to the command center.

03

Cyclone DDS: Proven High-Throughput Reliability

Performance pedigree: Eclipse Cyclone DDS is a battle-tested, high-performance DDS implementation known for its exceptional throughput and low latency on wired, high-speed networks. Reliability: It leverages the full DDS standard, including robust QoS policies (like RELIABLE and KEEP_LAST) that guarantee delivery semantics critical for industrial control. This matters for high-throughput, safety-critical multi-robot coordination where data loss is unacceptable, such as in factory-floor automation.

04

Cyclone DDS: Mature ROS 2 Ecosystem Integration

Default RMW: Cyclone DDS is the default RMW (ROS Middleware) implementation for ROS 2 Galactic and later, ensuring the broadest testing and community support within the ROS ecosystem. Tooling: It benefits from a rich set of established DDS tools for network monitoring, introspection, and security (DDS-Security). This matters for teams prioritizing a low-risk, well-supported path for production-grade ROS 2 deployments, where stability and community knowledge are paramount.

HEAD-TO-HEAD COMPARISON

Performance and Throughput Benchmarks

Direct comparison of key metrics and features for Zenoh and Cyclone DDS in multi-robot coordination over lossy networks.

MetricZenohCyclone DDS

Wire Overhead (per msg)

5-8 bytes

32-40 bytes

Peak Throughput (64-byte payload)

3.5M+ msgs/sec

1.2M msgs/sec

Latency (P2P, 99th %ile)

15 µs

30 µs

Wireless/Lossy Network Support

Peer-to-Peer Scaling

QoS Reliability (Reliable + Keep-Last)

Native ROS 2 RMW Support

Contender A Pros

Zenoh: Pros and Cons

Key strengths and trade-offs at a glance.

01

Minimal Wire Overhead & Peer-to-Peer Efficiency

Specific advantage: Zenoh achieves over 3.5M messages per second with a wire overhead of only ~5 bytes per data message, compared to the 30+ bytes typical of DDS RTPS. This matters for bandwidth-constrained teleoperation and drone swarms where every byte counts on lossy wireless links.

02

Unified Abstraction from Microcontroller to Cloud

Specific advantage: A single Zenoh protocol bridges deeply embedded MCUs (via zenoh-pico), ROS 2 robots, and cloud backends without protocol translation gateways. This matters for heterogeneous fleet architectures where you need a seamless data flow from a sensor on a Cortex-M4 to a dashboard in AWS, eliminating the integration tax of maintaining separate middleware stacks.

03

Superior Scalability for Dynamic Peer Discovery

Specific advantage: Zenoh's decentralized peer-to-peer architecture with multicast scouting allows nodes to discover each other without a central broker or complex XML-based discovery protocol. This matters for ad-hoc robot swarms and dynamic workcells where robots frequently join and leave the network, avoiding the discovery storms and single points of failure associated with DDS's Simple Discovery Protocol (SDP).

CHOOSE YOUR PRIORITY

When to Choose Zenoh vs Cyclone DDS

Cyclone DDS for Multi-Robot Fleets

Strengths: Proven reliability over lossy Wi-Fi networks, automatic discovery, and robust QoS policies (RELIABLE, KEEP_LAST) that are battle-tested in ROS 2 deployments. Cyclone DDS excels when coordinating heterogeneous AMRs in a warehouse where deterministic delivery and partition tolerance are non-negotiable.

Verdict: The safe choice for production-grade, multi-robot coordination where ROS 2 integration is a requirement.

Zenoh for Multi-Robot Fleets

Strengths: Minimal wire overhead and peer-to-peer scaling that bypasses the DDS discovery storm problem. Zenoh's zenoh-flow framework allows distributed data processing across robots without a centralized broker, reducing latency in dynamic fleet formations.

Verdict: Better for large-scale, dynamic fleets (50+ robots) where discovery traffic and bandwidth efficiency are primary concerns.

WIRE PROTOCOL SHOWDOWN

Technical Deep Dive: Protocol Architecture

A granular comparison of the wire protocol, discovery mechanisms, and architectural philosophies underpinning Zenoh and Cyclone DDS. This analysis targets lead robotics engineers evaluating middleware for high-throughput, lossy, and geo-distributed multi-robot systems.

Zenoh has significantly lower wire overhead. Zenoh uses a compact, binary encoding with a minimal header (typically 5-8 bytes), while Cyclone DDS adheres to the OMG RTPS (Real-Time Publish-Subscribe) protocol, which carries a larger submessage header structure (often 20+ bytes). For small, high-frequency messages common in robotics (e.g., joint states, IMU data), Zenoh's efficiency reduces bandwidth consumption by up to 80%. However, RTPS's verbosity includes built-in mechanisms for writer-side filtering and coherent presentation, which Zenoh handles at a higher layer.

THE ANALYSIS

Verdict

A data-driven decision framework for choosing between Zenoh's minimal overhead and Cyclone DDS's proven reliability.

Zenoh excels at minimizing wire overhead and scaling across diverse network topologies because its design decouples data exchange from rigid discovery protocols. In peer-to-peer benchmarking, Zenoh achieves over 3.5 million messages per second with a wire overhead of only 5 bytes per data message, making it exceptionally efficient for constrained networks like satellite links or large-scale geo-distributed systems. This minimalism translates directly to lower latency in wireless and multi-hop scenarios.

Cyclone DDS takes a different approach by fully embracing the Object Management Group (OMG) DDS standard, prioritizing deterministic reliability and rich Quality of Service (QoS). This results in a proven, battle-tested middleware for high-throughput, multi-robot coordination over lossy networks. In standard ROS 2 benchmarks, Cyclone DDS consistently demonstrates superior throughput for large messages, often saturating a 10GbE link with near-zero packet loss, a critical factor for transmitting dense point clouds and high-resolution sensor data between robots.

The key trade-off: If your priority is minimal network overhead, peer-to-peer scaling across WANs, and support for non-traditional transports like QUIC or Bluetooth, choose Zenoh. Its protocol-agnostic design is ideal for heterogeneous fleets and cloud-to-edge communication. If you prioritize proven, high-throughput reliability with strict real-time QoS guarantees over standard UDP multicast, choose Cyclone DDS. It remains the safer, more predictable backbone for a tightly coupled swarm of robots operating on a shared local network.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.