Eclipse iceoryx™

iceoryx2, the zero‑copy data plane

Move high-volume data efficiently and deterministically inside data-intensive systems.

Without moving the payload.

Read the book Get started on GitHub

Open sourceEclipse FoundationApache 2.0MIT

PerceptionPlanningControlActuatorsLiDARCamera

iceoryx2 zero-copy
data plane

Why iceoryx2?

For when moving data is the bottleneck.

In physical AI and other mission-critical systems, how data moves often becomes the deciding factor — long before the workload itself reaches its limit.

Cost of copying data

Data-intensive by nature

Data streams continuously, at high volume, between components.

Movement sets the limits

How that data moves determines latency, CPU usage, and predictability.

Copies don’t scale

Typical approaches copy and serialize payloads; overhead grows with size and message rate.

What it is

Built for performance, determinism, and control.

While iceoryx2 has its roots in zero-copy IPC, its role is broader: enabling efficient and predictable high-volume data exchange beyond process boundaries, without sacrificing architectural flexibility

Deterministic performance that scales

A true zero-copy data plane delivering deterministic performance, faster reaction times, and more CPU resources for your workload.

Designed for mission-critical systems

A communication foundation engineered for mission-critical systems, with predictable behavior and transparent operation needed for demanding environments.

Architectural sovereignty

A lean data plane that fits into your architecture instead of taking it over — no forced framework, no forced data model, no ecosystem lock-in.

How it works

One write. Many readers. Zero copies.

Typical communication mechanisms move data by copying payloads. iceoryx2 keeps payloads in place and exchanges references instead.

Producers Payload References Consumers

01
Produce

A producer borrows a slice of shared memory and writes its message directly into it. Once.
02
Share

iceoryx2 hands every consumer a reference to that exact memory—not a duplicate of it.
03
Consume

Consumers read in place. Adding consumers means sharing additional references, not payload copies.