Zettalane Systems participated to the recent The IT Press Tour in California a few days ago week and we learnt a lot about thair file and block storage approach leveraging ZFS and NVMe-oF.
Zettalane Systems is a cloud-native storage company focused on dramatically reducing public cloud storage costs while delivering high performance, simplicity, and enterprise-grade reliability. Founded by storage veteran Supramani “Sam” Sammandam, Zettalane builds on decades of experience in software-defined storage, ZFS, and NVMe technologies to rethink how file and block storage should be delivered in modern cloud environments. The company’s mission is to provide easy-to-consume, scalable, and cost-efficient storage using object storage and ephemeral NVMe resources, targeting up to 70% cost savings compared to traditional cloud storage services.
Zettalane positions itself against the high cost and architectural limitations of managed cloud NAS and block storage services. Traditional offerings such as AWS EFS or cloud block volumes are expensive, scale inefficiently, and often suffer from per-client throughput bottlenecks. Zettalane’s approach is to deliver cloud-native storage that exploits the massive bandwidth, durability, and scalability of object storage while preserving familiar NAS and block interfaces such as NFS, SMB, iSCSI, and NVMe-oF.
The company offers two primary products. MayaNAS is a high-throughput network file system designed for AI/ML, media, analytics, software development, and backup workloads. It uses a hybrid ZFS architecture in which metadata and small I/O are stored on local NVMe via ZFS “special vdevs,” while large sequential data blocks are written directly to object storage. This design avoids data tiering or migration and enables a single filesystem to handle both small-file IOPS and large streaming workloads efficiently. A key innovation is objbacker.io, a native ZFS object storage vdev that bypasses FUSE and enables highly parallel, direct I/O to cloud object storage using vendor SDKs. Validated benchmarks demonstrate over 8 GB/s read throughput in active-active HA configurations.
MayaScale addresses ultra-low-latency storage needs using local NVMe SSDs and NVMe-over-Fabrics. It provides both block and file modes, supporting databases, analytics, Kubernetes persistent volumes, and FSx-like workloads on clouds that lack native ZFS offerings. MayaScale uses server-side RAID-1 mirroring in active-active configurations, eliminating client-side replication overhead, reducing network traffic, and improving latency. Benchmarks show up to 2.3 million IOPS and sub-200µs latency, with future support for RDMA to further reduce latency where available.
Zettalane emphasizes automation and developer-friendly operations. Deployments are fully infrastructure-as-code driven using Terraform, enabling one-click provisioning, consistent multi-cloud behavior across AWS, Azure, and GCP, and rapid teardown for ephemeral workloads. The platform runs entirely within customer VPCs with no call-home requirement, addressing security and data sovereignty concerns. The business model is consumption-based, priced per vCPU with no per-GB or per-IOPS fees, and products are available through major cloud marketplaces.
Overall, Zettalane positions itself as a modern, cloud-first storage platform that bridges the gap between low-cost object storage and high-performance NAS and block storage, targeting cost-sensitive, performance-driven workloads with minimal operational complexity.
0 commentaires:
Post a Comment