sakkmesterke - Fotolia
Two of these are HPC primary storage with parallel file systems and RAM-equipped HDDs to speed reads and writes, while the other forms an archive tier with object storage and 8TB shingled magnetic recording (SMR) HDDs.
The Clusterstor G200 is a primary storage array with IBM's Spectrum Scale file system, formerly known as GPFS. Meanwhile, the Clusterstor L300 uses the Lustre parallel file system.
Both arrays are populated with Seagate Clusterstor HPC hard drives which are SAS-connected 4TB drives with RAM cache on board each drive. This is used to speed reads and writes. Seagate claims 300MBps throughput.
Clusterstor systems comprise controller technology from Seagate's Xyratex acquisition that form so-called Scalable Storage Units that come in 5U nodes that hold 84 drives.
Total capacities for Clusterstor clusters reach beyond 100PB with throughput of more than 1.6TBps possible, said Seagate storage systems vice-president Ken Claffey.
Drives can all be HDD HPC drives or alternatively two of these can be up to 800GB flash. An alternate capacity-led configuration allows all drives to be 8TB 7,200rpm SAS. Currently, the L300 can also be all-flash, and this variant will be offered for the G200 in 2016.
HPC storage systems need to provide file-based access with high capacity, throughput and I/O capability. They do this with high-scalable clustered NAS with parallel file systems. Suplliers vary by how they provide more rapid access and throughput, such as with the choices of performance-enhanced drive that Seagate offers.
Read more about high-performance computing storage
- We survey the key vendors in HPC storage, where huge amounts of IOPS, clustering, parallel file systems and custom silicon provide storage for massive number crunching operations
- 100,000 Genomes Project rejects building open-source parallel file system on x86 servers and opts for EMC Isilon clustered NAS deployed by Capita S3
Meanwhile, the Clusterstor A200 offers a bulk tier of archive storage with 8TB SMR drives, also in the 5U/84 drive configuration. Where the two primary storage arrays have dual storage servers and higher performance CPUs and memory, the A200 offers less in terms of performance.
In SMR drives data tracks do not sit next to each other as in standard HDDs, but are overlaid to increase capacity. This does, however, bring a performance penalty, particularly on writes.
The A200 stores data via a Seagate-developed object storage scheme but can be integrated with the G200 and L300 parallel file systems to provide an automated archive tier, via IBM Tivoli Storage Manager for the G200 and via the CEA-developed Robinhood rules engine for the Lustre-based L300.