A Universal Datastore For All Your Data

VAST software is deployed as stateless containers (CNodes) that are designed to be upgraded on the fly. These stateless machines are the logic of the cluster, executing all I/O, database and storage controller functions. DASE’s cache-less architecture means there is no data that needs to be failed over in the event of a container failure. VAST software runs in Linux containers, making it easy to upgrade CNodes without OS reboots. And every VAST enclosure that houses NVMe SSDs is fully redundant, never going offline during upgrades.

VAST’s asymmetric cluster architecture is designed to bring an end to the days of endless storage migrations. Clusters can scale over time using various generations of equipment where customers never need to think about or worry about where their data is…enabling an Infinite Cluster Lifecycle.

A system that grows infinitely needs to stay online indefinitely, so we designed this scale-out architecture to be online expandable. Customers can continue to add infrastructure as they need, and an online background process will rebalance metadata instantaneously while gradually rebalancing data in order to preserve precious flash write cycles.

The era of big data, big content and deep learning will require massive amounts of storage computing power. Historically, a scale-out cluster with more servers only experienced more failures, but the opposite is true with VAST.

With VAST's DASE architecture, each stateless CNode container can handle any cluster function, making it possible to survive the failure of all but a single machine and all while keeping the entire cluster online.

VAST's Locally Decodable Erasure Codes require only 2.7% data protection (parity) overhead at scale, playing a vital role in our mission to make all-flash more affordable for organizations of all sizes. This approach not only reduces costs but also enhances system resilience. Thanks to our n+4 redundancy, our clusters can handle twice as many failures as competing n+2 systems. During rebuild events, our new Locally-Decodable erasure codes only need to read from a fraction of a RAID stripe during a failure event, providing significant acceleration vs legacy Reed-Solomon-based approaches to data protection.  Additionally, data and parity are spread across all drives, enabling every cluster CPU and every NVMe drive to perform an accelerated rebuild event for peak performance.

Customers with a large number of storage enclosures choose to protect themselves from larger, rack-level, failures by enabling an additional safeguard we call DBox RAID. This new approach complements the fact that enclosures themselves are no-single-point-of-failure, and provides an additional safeguard even against the failure of an entire enclosure or even an entire rack. With DBox RAID, large systems can be architected for 99.999999999% uptime.