Fungible aims to make high-performance disaggregated storage attainable for data centers of any size with the launch of the Fungible Storage Cluster, the first in a series of data-centric platforms planned by the company.

The Fungible FS1600 Storage node is a two-rack-unit appliance that boasts 24 NVMe SSDs. However, what sets the platform apart from the more traditional storage servers are the pair of data processing units (DPUs) sitting at its core. Fungible announced its DPUs at Hot Chips 2020 back in August.

According to Fungible, these DPUs enable the FS1600 to achieve up to 15 million input/output operations per second (IOPS) over NVMe over transmission control protocol (TCP). In a 40-rack-unit chassis, that performance can scale to 300 million IOPS and beyond with additional racks, the company claims.

Fungible is the only vendor doing true software-defined storage today, according to Benny Siman-Tov, VP of product and business development at Fungible.

"What Fungible has done, for the first time in the industry, for data storage, is what SDN did for networking," he said in an interview with SDxCentral.

By separating the storage plane running on FS1600 from the control plane running on the Fungible Composer, an x86-based server, Fungible has been able to circumvent many of the bottlenecks that plague disaggregated storage, particularly when it comes to scale and performance, Siman-Tov explained.

Every other disaggregated storage platform has integrated the storage and control planes on the node, leading to challenges with scalability, according to Siman-Tov.

"As they plug in more and more notes, the control plane on each node has to talk to the control plane on all of the other nodes, and because of that, they get scalability issues," he said, adding that while x86 chips are great for single-threaded, high-performance workloads, they're poor at "data-centric processing."

The FS1600 doesn't have this problem because all data processing is done on the integrated DPUs, while the control plane is free to communicate across the nodes.

The Fungible Storage Cluster is designed to compute over NVMe-over-TCP to customers' compute servers running standard Ethernet interfaces.

This is ideal because customers don't need to invest in specialized interface cards to take advantage of Fungible's technology, according to Siman-Tov.

"The value that we bring is we are doing it over TCP," he said. "Traditionally, TCP is low performance, but with a DPU, we are getting... five-to-six times higher performance than fully customized storage platforms that are based on x86."

The Fungible Storage Cluster is available starting today.

Siman-Tov notes that the DPU's potential extends well beyond storage.

The company aims to make composable infrastructure a reality, enabling customers to define their resources based on the workload, not the confines of a single server.

The company's vision has helped it secure more than $300 million in funding from venture capitalists and investors, including Battery Ventures, Mayfield Fund, Redline Capital, and Walden Riverwood Ventures. The company raked in $200 million in its latest funding round in June.

Meanwhile, the company's DPU has won it endorsements from industry giants like Alibaba Cloud, Juniper Networks, and Samsung Electronics.