DataDirect Networks Inc., a mainstay of high-performance computing (HPC) storage trying to muscle into more mainstream environments, today unveiled a storage system that scales to more than 1 PB and supports 20 Gbps InfiniBand host connections.
The SA2A9900 StorageScaler, launched at the Supercomputing 2007 show at Reno, Nev., is a 4u configuration that holds 60 bays (up from 48 in its previous system) and 1,200 SATA or SAS drives in two data center racks. DataDirect said the SA2A9900 also performs reads and writes at 6 GBps (most systems write at half the speed as they read), supports up to 1,024 LUNs and RAID 6 and will offer 8 Gbps Fibre Channel hosts in early 2008.
The system also includes features from its predecessor, the SA2A9500, such as MAID and the use of heuristics and journaling to restore unresponsive drives. DataDirect also claims to sustain drive failures without slowing performance.
Dealing mostly with HPC environments so far, DataDirect claims 3,000 array sales totaling more than 100 PB, including 19 PB last quarter. Argonne National Laboratory, the U.S. Department of Energy's research laboratory, purchased 17 of the new SA2A9900 StorageScalers with more than 8 PB of SATA drives as the storage for a 556-teraflop supercomputer it is building.
One financial analyst with knowledge of DataDirect said the company has been profitable for several years and has already picked bankers to take it public in 2008.
Still, DataDirect is quiet compared to other storage companies that have recently filed for IPOs or are planning to, such as Compellent Technologies, Isilon Systems Inc., 3PARdata Inc. and BlueArc Corp. And, it's barely a blip on the radar screen compared to major storage vendors EMC Corp., Network Appliance Inc. (NetApp), IBM, Hewlett-Packard Co. (HP) and Hitachi Data Systems (HDS).
""They're a well-kept secret in storage," said Enterprise Strategy Group analyst Tony Asaro. "They haven't really reached out to the general IT market. They're focusing on a niche market. The challenge for them is to take that technology and adopt it to the mainstream market."
Asaro said DataDirect takes a different path than its rival storage vendors. While others are trying to differentiate themselves through software applications, DataDirect keeps beefing up its hardware capabilities.
"Everybody else is focusing on snapshots, replication and other software, and that's almost become the commodity these days," Asaro said. "DataDirect has really architected and optimized a system that can achieve numbers nobody else can."
Still, DataDirect won't be able to compete with the major storage vendors as a primary corporate storage system without the data protection capabilities, so it will have to catch on as an archiving system or primary system in rich media environments.
Asaro said capacity is a good selling point to break into its new target markets, with a caveat: "People want lots and lots of capacity. But it has to work."
An SA2A9900 system will have to work especially well to justify its pricing: Goldstein said the new StorageScaler will start in the $400,000 range and can go up to $2 million.
The Argonne Leadership Computing Facility (ALCF), which runs the supercomputers at the Argonne, Ill. lab, plans to use DataDirect's SA2A9900 as the main disk system for its 556-teraflops IBM BlueGene/P supercomputer. The lab is setting up the system, which also includes Myricom high-speed switches, and Argonne Leadership Computing director Ray Bair said it should be completed around mid-2008.
Bair said he chose the SA2A9900 because the lab has used DataDirect systems for five years and he is happy with the performance. He said speed is the most important factor, with reliability also key.
"A machine this big (556 teraflops) can do a lot of computation and will product a lot of data – a few hundred terabytes in a few days," he said. "We want to efficiently move that information to and from the BlueGene system."
The BlueGene/P runs without Fibre Channel. ALCF connects its file servers directly to the SAN through InfiniBand. The only switch uses is between the file server and BlueGene/P.