McCarony - Fotolia
Seagate Technology LLC Tuesday expanded its high-performance computing (HPC) portfolio with the introduction of a higher performing Lustre-based storage system, its first product supporting IBM's Spectrum Scale file system, and a new tiered archive object store.
The Lustre-based ClusterStor L300 and the G200, which is based on IBM Spectrum Scale and General Parallel File System (GPFS), include the new 3.5-inch hard disk drive (HDD) that Seagate's HPC systems group and drive team designed specifically for HPC applications. The Seagate storage systems are due in the first quarter of 2016.
The new ClusterStor HPC drive can store up to 4 TB in a single drive slot. Seagate storage claims the drive can deliver a sequential data rate of 300 MB per second and 35% faster random performance than any other 3.5-inch HDD. Distinguishing characteristics include the large cache and a caching mechanism that leverages some of the NAND flash within the HPC drive, according to Ken Claffey, vice president and general manager of Seagate's HPC systems business.
"We can say categorically it's the fastest disk drive that Seagate's ever made," Claffey said. "There is no other drive that comes close to those kinds of numbers."
The L300 and G200 can also use Seagate's solid-state hybrid drives or multilevel cell (MLC) SAS solid-state drives (SSDs) to boost performance. In an all-flash configuration, or "extreme mode," the ClusterStor systems can deliver up to 448 GB per second, per rack, according to Claffey. He estimated that only about 10% of the company's customers might be interested in the extreme mode, "where you need the most performance you can absolutely get, but not necessarily a lot of capacity."
Seagate's ClusterStor scale-out storage product family -- which also includes the 9000, 1500, Secure Data Appliance and Hadoop Workflow Accelerator -- is based on technology Seagate acquired last year from U.K.-based Xyratex Ltd.
The new L300 Seagate storage is the "bigger, badder, faster, better" follow-on to the ClusterStor 9000, which Seagate introduced in 2014, according to Claffey. Seagate devised a different letter and three-digit product nomenclature with the introduction of the Lustre- and GPFS-based systems, as well as the object-based A200 archive product at this week's Supercomputing 2015 conference in Austin, Texas.
Claffey said the L300 features upgrades to next-generation SAS, Intel CPUs, memory and interconnects, whether Intel's Omni-Path or Mellanox's Enhanced Data Rate InfiniBand. When equipped with the ClusterStor HPC drives, the L300 can supply up to 112 GB per second throughput per rack -- a 77% improvement over prior ClusterStor systems, according to Seagate.
The ClusterStor L300 supports object storage target disk pool management to create discrete storage pools for each type of storage media, whether flash, ClusterStor HPC drives, solid-state hybrid drives or high-capacity drives.
Seagate storage extends support to GPFS
The ClusterStor G200 is the result of a partnership Seagate forged over the summer with IBM. The custom-designed, preconfigured, GPFS-based storage system has 84 slots for 3.5-inch HDDs and SSDs in a 5U enclosure with embedded storage controllers. Seagate said ClusterStor for IBM Spectrum Scale uses Grid RAID, Seagate's declustered RAID to enable rebuild times up to 400% faster than legacy RAID systems.
The most popular parallel file systems among HPC users are open source Lustre and IBM's GPFS, with roughly an even split between the two. Most major storage vendors serving the HPC market offer both in order to exploit more of the market, according to Steve Conway, research vice president for HPC at IDC, based in Framingham, Mass.
Addison Snell, CEO of Intersect360 Research in Sunnyvale, Calif., said open source Lustre is less expensive to license than GPFS, and more common in academia and the public sector, whereas GPFS is more common in commercial deployments -- such as finance, oil and gas, and manufacturing.
New A200 object store
Seagate's new HPC-focused A200 object store was designed to complement Seagate's Lustre- and GPFS-based storage systems. The active archive system incorporates newly developed Seagate software that lets customers migrate data from higher priced primary storage for fast retrieval.
He said the A200 storage system uses high-capacity, power-efficient shingled magnetic recording-based disk drives. The A200 consists of up to seven scalable storage units per rack, with each unit capable of storing 524 TB for a possible total usable capacity of 3.63 PB per rack. Additional storage units and racks can be added to the pool. Included with the purchase of ClusterStor products is a management suite that customers can use across any of the storage systems.
The A200 system uses third-generation ClusterStor erasure coding for data protection. The network-based erasure code can use the resources of the entire system to achieve fast rebuilds of failed drives and can prioritize critical data elements to rebuild, according to Seagate.
Snell said Intersect360's most recent user survey showed that only approximately 20% of HPC users currently have object storage, but that percentage is rising. He said object storage's advantages for HPC include the ability to use metadata to describe, manage and access data.
"There's a whole bunch of different use cases out there. And to me, it looks like [Seagate is] trying to essentially cover the waterfront," said Bob Sorenson, a research vice president in IDC's HPC group.
Seagate did not provide suggested list prices for the A200, G200 and L300. The company sells ClusterStor products through a network of OEMs and resellers, including Bull, Cray, Dell, HPE and SGI.
Seagate extends flash storage product line
Kinetic object storage from Seagate unveiled
Seagate backup, cloud strategy expands with EVault backup appliance