Cloud Native 10 min read

Nydus: Open‑Source Container Image Acceleration Service for Faster and Secure Deployments

The article introduces Nydus, an open‑source container image acceleration service co‑developed by Ant Group and Alibaba Cloud, explaining its architecture, on‑demand download, data deduplication, end‑to‑end consistency, OCI compatibility, and how it speeds up container startup while enhancing security.

AntTech
AntTech
AntTech
Nydus: Open‑Source Container Image Acceleration Service for Faster and Secure Deployments

Ant Group and Alibaba Cloud have jointly open‑sourced the container image acceleration service Nydus, which is now part of the CNCF‑incubated Dragonfly project, aiming to build a richer cloud‑native ecosystem.

In production, small container images deploy quickly, but images that are several gigabytes take a long time to download to a node. While Dragonfly improves large‑scale distribution via P2P, users still must wait for the full image to download before a container can start. Nydus addresses this by dramatically reducing download time and providing end‑to‑end data integrity verification.

Key Features of Nydus

On‑demand image download – containers can start without the entire image being present locally.

Block‑level data deduplication – saves storage resources.

Only the data needed for execution is stored; obsolete data is not kept.

End‑to‑end data consistency checks for better data protection.

Compatibility with OCI distribution and artifacts standards, ready to use out of the box.

Support for multiple storage backends (image registries, NAS, S3‑compatible object stores, etc.).

Seamless integration with Dragonfly.

Architecture

Nydus consists of a new image format and a userspace FUSE filesystem process that parses container images. The format separates metadata and data layers; the metadata layer is a self‑checking hash tree where each file and directory is a node with its own hash. File data is sliced into fixed‑size chunks that can be shared across files and images, enabling deduplication.

The FUSE (or virtiofs) filesystem allows Nydus to serve both traditional runc containers and Kata containers. Image sources can be container registries, OSS object storage, NAS, or Dragonfly super‑nodes and peers. A local cache can be configured to avoid repeated remote fetches.

Benefits to Users

Deploying Nydus makes container startup noticeably faster—common images that previously took minutes now start in seconds. Additionally, because Nydus does not fully extract images to the local filesystem, it can verify data on each access; if tampering is detected, the data is re‑fetched from the remote source, providing stronger runtime integrity.

Future Plans

After a year of internal stabilization, security hardening, and usability improvements, the project is open‑sourced. The vision is that when users run Dragonfly and Nydus together, container applications can be launched quickly regardless of image size, without concerns about image data safety.

OCI Community Involvement

The team actively participates in OCI discussions about the next‑generation image format (OCIv2). They propose Nydus as a reference implementation because it satisfies all listed OCIv2 requirements, such as reduced duplicate data, reconstructable format, minimal filesystem metadata, mountable filesystem, on‑demand loading, and verifiable content.

FAQ

What problems exist with the current OCI image standard? It relies on the legacy tar format, which is outdated for container images.

How does Nydus differ from CRFS? Both share similar design ideas, but Nydus adds block‑level deduplication and end‑to‑end consistency checks, extending the CRFS stargz format.

How does Nydus compare to Azure Teleport? Teleport still uses the tar‑based OCI format, while Nydus abandons tar in favor of a merkle‑tree format with advanced features.

What happens if the network drops while a Nydus‑based container is running? Containers can start before the full image is downloaded; after startup, Nydus continues background downloading, so a network interruption does not prevent the container from running once the needed data is cached.

Appendix: OCIv2 Image Standard Requirements

Since June 2020, the OCI community has discussed shortcomings of the current spec and defined requirements for the next‑generation format, including less duplicate data, reconstructable images, reduced filesystem metadata, mountable filesystem, on‑demand loading, extensibility, verifiable checksums, lower upload bandwidth, and operation on untrusted storage. Nydus meets all these criteria.

For the full list of requirements and detailed discussion, see the shared document: https://hackmd.io/@cyphar/ociv2-brainstorm

— END —

cloud-nativecontainerFilesystemdragonflyNydusimage accelerationOCI
AntTech
Written by

AntTech

Technology is the core driver of Ant's future creation.

0 followers
Reader feedback

How this landed with the community

login Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.