As part of a comprehensive product line refresh and rollout that will extend through SC next week, DataDirect Networks (DDN) today launched the next generation of its high performance SFA block storage line with SFA14K and SFA14KE, formerly codenamed “Wolfcreek.” No stranger to high performance – the company supports over two-thirds of the world’s 100 fastest supercomputers – DDN is betting big that the next generation interconnect technologies are far enough along to incorporate in the new offering and attract performance-seeking customers.
“We could have launched the products six months ago,” said Molly Rector, DDN chief marketing officer, “but we didn’t want to release them without EDR which wasn’t yet available.” Built for speed, the new appliances have a built-in PCIe fabric to speed data within the device. Support for NVMe, EDR/IB, 100Gb Ethernet, and OmniPath Fabric are all included (see full specs below).
Rector declined to discuss pricing but like all block storage, the offerings are built for performance. DDN says SFA14K and SFA14KE deliver the world’s fastest, densest storage with more than 6 million IOPS and 60GB/sec in 4U with scalability to over 7PB of capacity in a single rack. The new products, said Rector, will bridge in advanced interconnect technologies now coming online in the effort to push towards exascale computing.
DDN has its eyes on two distinct markets, traditional HPC, where it has long been a dominant player, and the emerging enterprise market for advanced scale computing and hyper-converged datacenters. DDN characterizes the SFA14K as a hybrid NVMe SSD and disk block storage array. Its near twin, the SFA14KE, is a hyper-converged solution that leverages the embedded processor to run virtual machines, embedded applications and file systems within the storage array to significantly reduce complexity, latency and datacenter footprint.
The SFA14K family is a proprietary vertically integrated system intended in part to deal with the data avalanche in science and industry. By designing it from the ground up and using custom hardware (and software), DDN has reduced the number of components typically found in large commodity-based storage systems. One result, says the company, is dramatically reduced latency.
Rector said, “There’s been a shift of the last few years to use ‘white box’ commodity hardware [in large storage systems].” The do-it yourself approach produces purchasing flexibility, she agreed, but with a substantial performance penalty, especially as installations are expanded. Latency problems can become extremely problematic as more devices are incorporated, argued Rector.
The challenge, according to Rector, is how to integrate all these moving parts in an optimal way. “Every time you add a new box, you have more ports, more switches, more cables, more software applications to manage,” she said, which can hobble throughput in performance-sensitive settings. The SFA14K line dramatically reduces latency because everything is designed to work together – a distinct advantage as datasets grow larger, she said.
“The core of [SFA14K] is the really high-performance PCIe fabric that is designed so every single device in the system can be run at its highest performance specified speed,” Rector emphasized. “We can put either 48 NVMe devices or 72 SAS SSD devices in the controller chassis. Then we can run all of them at line speed at 100 percent spec’d performance.”
The SFA14K/14KE products are a significant step up from the SFA12K generation which runs at 48Gb/s and just over 4M IOPS and doesn’t have the embedded PCIe fabric. DDN presented SFA14K’s impressive benchmarks against all flash devices from Pure Storage and EMC, although without much detail on the specific tests run.
DDN cites a variety of benefits:
- 10x reduction in network\ports, cables, HBAs
- 10-100x latency reduction
- Simple upgrade path to new technology: NVMe; 100Gb/E; EDR Infiniband; Omni-Path; new applications
- 50-70% reduced IT headcount to manage
Broadly, DDN says the DDN SFA14K family value proposition is: “Data intensive enterprise and HPC organizations are requiring a groundbreaking new approach to data compute, storage and management that allows them to keep pace with the rapid acceleration of processor power and both machine and sensor generated data. Data is being created at a staggering rate and needs to be stored efficiently while being quickly available to analyze and leverage for strategic decision making. Today’s organizations need new architectural designs that can support the full power of advanced individual component technologies with efficiency and integration at scale.”
Rector said, “While many storage vendors have discontinued investing in hardware innovation, DDN consistently executes upon a product strategy to break through bottlenecks in the datacenter. And while DDN has embraced the movement to deliver software technology on commodity hardware, we also recognize that commodity storage platforms simply cannot keep pace with rapidly increasing compute power, network bandwidth and the latest generation NVMe and SSD speeds.”
Serving both the HPC and high enterprise markets gives DDN a good vantage point from which to see differing trends. Ongoing adoption of parallel file systems – primarily Lustre and GPFS (IBM) – remains strong in both markets, said Rector. Interestingly, Lustre is stronger in Europe than the US, she said, where GPFS’s more extensive feature set has given it an edge. “But Lustre is catching up,” she said.
OpenStack adoption and performance related worries are growing in the HPC community, according to Rector, while Hadoop, VMware are the big care-abouts in the enterprise. Although it must be said the virtual machine/container technology is starting to garner attention in the HPC world as well. NERSC, for example, recently released Shifter, a software tool for bringing Docker containers onto Edison, NERSC’s Cray Supercomputer.
DDN says the new SFA14K/KE product will ship this month.