TL;DR
At the OpenZFS Developer Summit 2025, Zettalane presented MayaNAS and MayaScale alongside objbacker.io, a native ZFS VDEV implementation that avoids FUSE and maps large ZFS blocks directly to cloud object stores. Benchmarks on AWS reported up to 3.7 GB/s sequential reads and 2.5 GB/s sequential writes using a multi-bucket striping approach.
What happened
Zettalane presented its MayaNAS file-storage and MayaScale block-storage platforms at the OpenZFS Developer Summit 2025, highlighting objbacker.io — a native ZFS VDEV (VDEV_OBJBACKER) that communicates with a userspace daemon through /dev/zfs_objbacker. The daemon, written in Go, uses cloud provider SDKs to issue object PUT/GET/DELETE operations that correspond to ZFS I/O types, removing the double kernel↔userspace crossings typical of FUSE mounts. With ZFS recordsize set to 1 MB, each ZFS block can map to a single object, enabling large aligned transfers to the object store. Performance tests on AWS c5n.9xlarge instances (36 vCPUs, 96 GB RAM, 50 Gbps) showed up to 3.7 GB/s sequential read and 2.5 GB/s sequential write, achieved by striping across six S3 buckets to parallelize I/O. The presentation also covered MayaScale's NVMe-oF blocks and multi-cloud deployment patterns.
Why it matters
- Bypassing FUSE reduces context-switch overhead and can improve throughput for large sequential workloads.
- Mapping large ZFS blocks directly to object storage enables a hybrid tiering model: local NVMe for metadata/small IOPS, object stores for high-throughput data.
- Parallelizing across multiple buckets can saturate available network bandwidth and lift cloud object storage performance toward block-store levels for certain workloads.
- If broadly adopted, native VDEV integration could lower cloud NAS costs; the presenter reported 70%+ savings versus traditional cloud block storage.
Key facts
- Presentation venue: OpenZFS Developer Summit 2025, Portland, Oregon.
- Component: objbacker.io implements a native ZFS VDEV type named VDEV_OBJBACKER and exposes /dev/zfs_objbacker.
- Architecture: userspace Go daemon with frontend for ZFS VDEV and backend using AWS, GCP, and Azure SDKs.
- ZIO mapping: WRITE→PUT, READ→GET, TRIM→DELETE, IOCTL (sync)→flush operations to object storage.
- Data alignment: with ZFS recordsize 1 MB, each block maps to a single object (aligned writes send direct PUTs).
- Benchmark platform: AWS c5n.9xlarge (36 vCPUs, 96 GB RAM, 50 Gbps network).
- Measured throughput: 3.7 GB/s sequential read from S3 and 2.5 GB/s sequential write to S3.
- Parallelism: tests used six S3 buckets configured as a striped pool to parallelize I/O across endpoints.
- FIO test details: 1 MB recordsize, 1 MB block size, 10 concurrent jobs, 10 GB file per job (100 GB total), sync I/O engine.
- MayaScale: NVMe-oF block storage with Active-Active HA and multi-tier IOPS figures reported for GCP (various tiers).
What to watch next
- OpenZFS YouTube channel for the full 50-minute presentation video (to be published by OpenZFS).
- Whether objbacker.io or its VDEV design will be accepted upstream into mainline OpenZFS or distributed as an open-source project — not confirmed in the source.
- Long-term production validation, data durability/consistency results, and broader interoperability testing across providers — not confirmed in the source.
Quick glossary
- ZFS: A combined file system and logical volume manager known for data integrity features, snapshots, and flexible pooling.
- VDEV: A virtual device in ZFS that represents physical storage; pools are built from one or more VDEVs.
- FUSE: Filesystem in Userspace, a kernel module that allows filesystem implementations to run in user space, typically at the cost of extra context switches.
- Object storage: A storage architecture that manages data as discrete objects (with metadata and identifiers), commonly exposed by cloud services such as S3, GCS, and Azure Blob Storage.
- Recordsize: A ZFS filesystem parameter that defines the maximum size of ZFS blocks; alignment with object sizes affects I/O efficiency.
Reader FAQ
Does objbacker.io remove FUSE entirely?
Yes — the design replaces FUSE-mounted buckets with a native ZFS VDEV and a userspace daemon accessed via /dev/zfs_objbacker, avoiding the FUSE kernel↔userspace round trips.
What performance was demonstrated?
Benchmarks on AWS reported up to 3.7 GB/s sequential reads and 2.5 GB/s sequential writes using six striped S3 buckets on c5n.9xlarge instances.
Is objbacker.io included in mainline OpenZFS?
not confirmed in the source
Is the software open source or available for download?
not confirmed in the source
Home Blog MayaNAS MayaNASZFSArchitecturePerformanceAWSGCPAzure MayaNAS at OpenZFS Developer Summit 2025: Native Object Storage Integration with objbacker.io December 27, 2025 • 10 min read • Supramani Sammandam We presented MayaNAS and…
Sources
- Native ZFS VDEV for Object Storage (OpenZFS Summit)
- OpenZFS Developer Summit
- Klara at the OpenZFS User & Developer Summit 2025
- OpenZFS Developer Summit 2025
Related posts
- Samsung shifts Gaming Hub’s focus toward social and discovery tools
- Netflix rolls out its first original video podcasts with Pete Davidson, Michael Irvin
- Webctl: CLI browser automation for AI agents as alternative to MCP