Pavilion Data: Our storage array is considerably better than yours

The legacy design of dual-controller storage arrays is unsuited for today’s performance and capacity demands. So claims Pavilion Data, the NVMe-over-Fabrics array startup.

All suppliers will need to junk dual-controller systems if they are to cope with growing data volumes, CEO Gurpreet Singh told a press briefing last week.

Unfurling the company’s near-term roadmap, Singh said the company’s Hyperparallel Flash Array (HFA) deliver unmatched performance and capacity and represents a “third wave of computing”.

Pavilion Data says its HFA beats file and block access competitors and provide better throughout per rack unit than object storage competitors.

HFA has up to 20 controllers, with shared memory for metadata, and fast access to NVMe SSDs with parallel access to the SSDs. The system supports block, object and NFS file access. The company said SMB support is contingent on customer demand. Each controller is dynamically defined as a block, file or object access controller, and a second controller acts as a standby backup.

U.2 format SSDs only are supported. However the SSDs are mounted on carriers, so fresh formats could be supported by redesigned carrier cards.

Pavilion’s roadmap includes 30TB SSD support in the second half of this year along with an IO card and Optane SSD support.

V2.4 of the Pavilion OS will add a fast object store, Nagios system monitoring software integration, data compression and Windows drivers for NVMe-oF TCP and RoCE.


Pavilion likes to compare HFA performance using raw IOPS and GB/sec numbers with latencies, and add cabinet rack unit take-up and capacity to emphasis performance rack space density.

In the briefing, it contrasted a 4U, 1PB usable capacity HFA system for block access with publicly available numbers from a Dell EMC PowerMax (80U and 3PB), NetApp’s AF800 (48U and 4.4PB) and a Pure Storage Flash Array (6 to 9U – it was uncertain which – and 896TB)

The diagram shows Pavilion’s IOPS and GB/sec superiority:

This is a grab from a Pavilion slide with supplier labels added

This Pavilion slide compares file access with Pure Storage’s FlashBlade and Isilon and VAST Data systems.

This is a grab from a Pavilion slide, with supplier labels added

And it has repeated the exercise with Pure Storage, MinIO and OpenIO for object storage.

This is a grab from a Pavilion slide with supplier labels added.

Pavilion is slower in absolute throughput terms than MinIO and OpenIO but uses much less rack space. The company said: “Pavilion outperformed all competitors in the same rack space. When normalized to a per RU basis, we actually outperformed MinIO and OpenIO as well.”

 It outperformed Pure in the same rack space. The roadmap includes a fast object store later this year and this may swing the object performance meter in Pavilion’s favour.


Gurpreet Singh said Pavilion Data has broken free from the other NVMe-oF array startups (we think he is referring to Excelero), citing last year’s $25m capital raise which takes total funding to $58m. The company has more than 85 employees and “many” customers, though it won’t say how many. It claims one US Federal customer runs the world’s largest NVMe-oF system but it can’t reveal the name, nor the size of the system.

Public customers include the Texas Advanced Computing Center (TACC), where three Pavilion HFAs replaced five EMC DSSDs, and Statistics Netherlands.