ZettaLane Claims 13 GB/s Performance from Google Cloud Storage Using ZFS Instead of FUSE
Key Takeaways
- ▸ZettaLane's MayaNAS achieves 13 GB/s throughput from $0.02/GB Google Cloud Storage using ZFS instead of FUSE
- ▸FUSE-based object storage mounts suffer from fundamental limitations including 83% performance degradation, no hard links, non-atomic renames, and whole-object rewrites
- ▸MayaNAS uses a hybrid architecture with ZFS treating object storage buckets as block devices, combined with local NVMe for metadata
Summary
ZettaLane Systems has published a technical blog post challenging the conventional approach to bridging object storage and POSIX filesystems in cloud environments. The company's MayaNAS product uses ZFS (Zettable File System) as a kernel-level filesystem layer on top of object storage, claiming to achieve 13 GB/s throughput from Google Cloud Storage on a single VM—significantly outperforming traditional FUSE (Filesystem in Userspace) implementations.
The post extensively critiques FUSE-based solutions, which are commonly offered by major cloud providers to mount object storage buckets as filesystems. ZettaLane argues that FUSE's architecture creates insurmountable performance and functionality limitations, including up to 83% performance degradation, lack of hard links, non-atomic operations, and whole-object rewrites for any modification. The company cites Linus Torvalds' criticism of userspace filesystems and academic research from USENIX FAST '17 to support their position.
MayaNAS instead treats object storage buckets as block devices that ZFS can stripe across, combining them with local NVMe storage in a hybrid architecture. The solution is deployed via Terraform on major cloud platforms (GCP, AWS, Azure) or through cloud marketplaces. ZettaLane positions this as fundamentally solving the "POSIX bridge problem" by using a proper kernel filesystem rather than working around FUSE's limitations with faster backend infrastructure.
- The solution is available via Terraform deployment or cloud marketplaces for GCP, AWS, and Azure
Editorial Opinion
ZettaLane's critique of FUSE is technically sound—the architecture really does impose unavoidable overhead and POSIX incompatibilities. However, the 13 GB/s claim lacks crucial context: what workload pattern, what VM specifications, what cost comparison to managed alternatives? The company is clearly positioning against Google's own Filestore and Parallelstore offerings, but without third-party benchmarks or customer case studies, this reads more like a well-researched technical argument than proven disruption. The fundamental insight—using a real filesystem instead of userspace emulation—is correct, but enterprise adoption will depend on operational maturity, not just architectural elegance.
