Object storage supplier Cloudian has launched v8.0 of its HyperStore software with combined file and object access methods, a single namespace, high-performance and single point management for a geo-distributed set of Cloudian nodes. It’s looking at GPU servers having direct access to object storage.
Object storage began its technology life as a more scalable and simpler way of storing large data sets by abandoning the nested file:folder addressing structure of file systems and using a single-level addressing scheme based upon the objects’ contents. However file system scalability was increased with scale-out and also parallel access systems. At the same time object storage, with its simple access and scalability, became the default method used in hyperscale public cloud providers such as Amazon, Azure and Google. Inevitably the two data access method started being combined, with filesystem suppliers adding S3 support, and object storage suppliers adding NFS and CIFS/SMB access.
A statement from Cloudian CEO Michael Tso said: “The days of legacy storage systems, with their single-site storage silos, limited protocol support, and complex proprietary infrastructure are over.”
In his view: ”HyperStore 8 sets a new benchmark in performance and efficiency by enabling our customers to fully leverage their object and file data across both modern and legacy applications regardless of the data capacity, physical location, or data format, all on industry standard hardware and within a simple global management framework.”
Cloudian first started integrating file access technology in 2017 when it acquired Infinity Storage, an Italian business, and introduced its HyperFile software with file caching added to HyperStore. By July this year HyperFile had been developed into a NAS controller providing SMB/CIFS and NFS file services, with HyperStore object storage as its underlying storage layer.
Now HyperStore 8 integrates file and object access object storage seamlessly within Cloudian’s modular, software-defined-storage platform. Data is accessible in both file and object formats, regardless of its original format, making information available to both file and object-using applications.
Cloudian is also adding central point management of a geo-distributed fleet of HyperStore instances with app-aware policy management for local performance and capacity optimization, security and data sovereignty. HyperStore 8 has good all-flash performance with 17.7GBps writes and 25GBps reads from a six node all-flash cluster in a recent benchmark. Cloudian’s per-node performance was 2.95 GBps write and 4.15 Gbps read. Scale up node compute or scale out node numbers to get more performance.
File systems can employ Nvidia’s proprietary GPUDirect protocol to get direct access from Nvidia GPU servers to NVMe storage drives. There is no industry-standard equivalent direct access method between AMD and Intel GPUs and filesystems, or between any GPUs; AMD, Intel or Nvidia, and object storage systems.
Cloudian CMO Jon Toor told us: “The AWS S3 API does not currently support GPUDirect. Along with other stakeholders here, we are looking at several means of providing this direct access to storage while remaining compliant with the AWS S3 API. While we have not yet settled on a path, we definitely see the need and are excited by the possibilities here.”
The company says HyperStore has a rich S3 API compatibility, public cloud integration, military-grade security, and ransomware protection. Cloudian claims HyperStore 8’s capabilities makes it ideal for both capacity-intensive and performance-intensive workloads in AI/ML, data analytics, data protection, and hybrid cloud.
You can read the HyperStore 8 software-defined storage datasheet here and the HyperStore appliance datasheet here. Cloudian HyperStore 8 is available now for evaluation.