Accelerating Modern Data-Driven Applications

Reducing cost and complexity.

I had the opportunity to meet with Dario Zamarian, CEO, and Costa Hasapopoulos, Chief Field Technology Officer at Pavilion, the maker of the hyper-parallel flash array.

Pavilion's industry and solution focus are on life sciences, media and entertainment, financial, high-performance databases, high-performance virtualization, and federal as those industries are driving a new set of hyper-parallel data-centric applications. Hyper-parallel infrastructure provides scale-up and/or scale-out applications by delivering consistent, predictable, scalable, and low latency performance. Storage should also be hyper-parallel with the performance of DAS and all the benefits of shared storage.

Pavilion HyperOS is their foundational storage solution, a flexible and high-performance storage platform with enhancements to block, file, and object. This provides the flexibility of deployment using the client's file system or Pavilion's file, block, or object storage with best-in-class performance density. There are multi-client options and interoperability to provide the ultimate flexibility. The architecture is manageable as well as scalable.

The Pavilion platform uses an API-first approach and provides management and ecosystem integrations with Windows, VMware, and VMware Tanzu Cloud Foundry.

The architect is unique on several fronts. It is built like a network switch with a PCIe backplane, build from the ground up to support NVMe and NVMe-oF with multiple controllers and network ports. Connects via ethernet or InfiniBand. Works with any controller to any drive connectivity. It is cache-less and tier-less. DMA from any drive and any controller in nanoseconds. RDMA from hosts to Pavilion in microseconds. Software patents to take advantage of the hardware. And, ability to Tier outside of the platform to spinning disk.

According to Gartner:

Next-generation, data-centric workloads like artificial intelligence (AI), modeling, and simulation require bandwidth, latency, and input/output operatiuons per second (IOPs) performance beyond the capabilities of NVMe technology-enabled, dual-contraoller scale-up of all-flash storage arrays at scale.

According to Michael Kagan, CTO at Nvidia:

High performance computing requires high performance IO. The higher the processing power the compouter element, the more data it can process hence faster deliver is required. Performance improvement for file storage of almost 4X by using GPU direct storage. Storage access is yet another bottleneck that needs to be resolved.

Pavilion solutions are being adopted in high data ingest, AI, and real-time decision support on large data sets. They are gaining sales the Federal sector with federal government and the intelligence community.

Specific projects include analytics workloads in financial services, cybersecurity, facial recognition, signals processing, and extreme science discovery. In media and entertainment there are projects around volumetric capture and post-production. In life sciences, projects include cryogenic electron microscopy, genomics, and radiology. They are also seeing uptake in virtualized infrastrcuture.