Vast Data will provide a containerised storage software version of its product. And that’s not too much of a stretch because the all-NVMe flash “universal storage” provider that claims to be the nemesis of the hard drive is indeed built on an architecture that includes a containerised control layer.
But the move to a containerised storage software product will not happen for some time because for now Vast is focussed on – in the grand scheme of things – smaller targets in enterprise storage.
We asked Vast – at its recent live event – whether the company had plans to offer the container component of its product as a software-only offering.
The move would make sense because Vast’s architecture is based on provision of QLC flash and 3D Xpoint storage hardware, with a software control plane comprising stateless Kubernetes containerised servers with a view of all the underlying storage. It follows, then, that the software control plane could be re-factored to other media, of the customer’s choosing perhaps.
“We have a lot of plans in this space,” said chief technology officer Sven Breuner. “But today we have specific hardware dependencies at relatively small scale. But we could build large infrastructures with commodity capacity.
“For now we want to focus on our specific platform, although some customers are already running Vast as a containerised deployment on bare metal as proofs-of-concept, but not many.”
Vast Data’s architecture rests on bulk, but rapid access storage provided by NVMe-connected QLC flash. QLC, is relatively cheap for flash, but is the least durable of all flash generations, and is best used for sequential input/output (I/O). To ensure traffic is sequentialised as much as possible Vast puts in a layer of 3D Xpoint to shape I/O into fewer, less randomised patterns.
According to Vast there is 18TB of XPoint (in 12 15.36TB drives) for every 675TB of QLC, making about 2.5% of the system by capacity.
The result of all this is a set of storage economics and use profiles that Vast claim is the death knell for the spinning disk hard drive and for tiered storage. It calls itself “universal storage” and “an extinction-level event for hard drive tiering”.
Despite being comprised of all-NVMe flash, Vast claims to offer storage cost per GB that is “within spitting distance” of HDD pricing and “less than the cost of the hardware”, according to CME Jeff Denworth. That’s because it uses QLC for the bulk of its capacity and by use of very granular data reduction – at the global namespace level – that packs more data into less space.
Vast is aiming to get customers to repatriate data from cloud storage and offers the promise of not having to deal with capacity management. It targets use cases that include analytics and AI, HPC, transactional processing, VMware workloads, container deployments.
Vast can provide storage for use cases in any sphere, “that does not need block storage primitives”, said Denworth, namely “SCSI persistent reservations” which is akin to file locking and allows only specified to access nodes to access a LUN.
Despite building a slogan around “democratising fast access to data”, Vast provides only file and object modes that include NFS, SMB and S3, but also direct access to memory via RDMA and container persistent storage via CSI drivers.
The company is happy to live with that for now, however.
“File and object are the best place to sell flash infrastructure,” said Denworth. “We are consciously not targeting the block storage market. File and object are the ‘easy’ button. Also, most data is on file or object storage and you can do most things with file.”