News Stay informed about the latest enterprise technology news and product updates.

Pavilion Data gets funded for NVMe-oF push

As all-flash array pioneer Pure Storage celebrates its $1 billion in annual revenues, two of its former executives are planning the next big thing in flash.

Pavilion Data, whose CEO Gurpreet Singh and VP of global sales Dan Heydenfeldt came from Pure, today picked up $12 million in funding to market its NVMe over Fabrics (NVMe-oF) storage system. Singh said Pavilion is going after customers running applications built on a “new modern stack dominated by open source, massively parallel, scale-out, clustered databases and file systems.” In other words, it’s targeting storage for databases such as MongoDB, Spark, MySQL and Cassandra instead of Oracle and Microsoft SQL Server.

“Somebody gets to build the next billion dollar company riding on this modern data stack,” Singh said. “We believe we have the best architecture for these modern applications. We call it disaggregated shared storage, or rack-scale flash.”

Singh said current all-flash arrays — including Pure’s — are fine for traditional applications but not modern apps. “The old-school dual controller architecture, server-centric design exposes a lot of challenges when running these applications,” he said. “For example, performance density is just not there.”

He said a storage system to run these new apps must have performance, latency and bandwidth characteristics of direct attached storage yet easy to scale and use. “Today there are compromises,” Singh said. “You can go shared storage, but you lose performance. Or you stick in four to six NVMe cards per server, 40 servers per rack, tens of racks and you lose the serviceability and data management. “It requires a complete re-think of how you develop and architect a storage system. You can’t retro-fit that. The basic math doesn’t add up.”

Pavilion’s answer is the Pavilion Memory Array, which started shipping in early 2018.

Pavilion Data CTO VR Satish said a 4u Pavilion Memory Array can drive 120 GBps second with around 100 microseconds of latency and 20 million IOPS. The system uses x86 hardware and up to 72 standard 2.5-inch NVMe flash drives for a maximum capacity of 1 PB.

The back of the box resembles a network switch with a minimum of two line cards, each card containing four 100 Gigabit Ethernet ports and two controllers. Customers can expand to 10 line cards, 40 GigE ports and 20 controllers.

The array does not scale beyond one system, but Satish said 1 PB of storage with 72 drives “is more than enough.”

The system supports RDMA over Converged Ethernet (RoCE) and NVMe over TCP but no Fibre Channel.

“Hyperscalers don’t want to be caught dead running Fibre Channel in their data centers,” Singh said.

Satish said Pavilion developed a clustered file system that supports RAID 6 data protection, non-disruptive upgrades, multi-pathing, thin provisioning, snapshots and clones. The array does not require any software to run on host servers.

Founded in 2014, Pavilion Data has just under 60 employees split between the U.S. and India. But the company has not had all smooth sailing so far. A founder and the original CEO Kiran Malwankar left Pavilion early this year. Pavilion also had some layoffs around that time, although it is hiring now.

Singh said another founder and current VP of engineering Sundar Kanthadai is in charge of development, and there has been no change in direction since Malwankar left.

He said Malwankar “left to pursue other opportunities” although Malwankar’s LinkedIn entry describes him as a “free bird.”

“It’s the natural course and evolution of a company,” Singh said. “People leave and new people come in.”

The new funding brings Pavilion Data’s total to $33 million. New investors Korea Investment Partners and DAG Ventures participated along with previous investors Kleiner Perkins Caufield & Byers, Artiman White Space Investments, and SK Telecom.

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.

-ADS BY GOOGLE

SearchDisasterRecovery

SearchDataBackup

SearchConvergedInfrastructure

Close