DataDirect, a shipper of very high-speed block-access storage to the high performance computing (HPC) and media worlds, is now offering native file access.
The company has seen the way two winds are blowing. The first is that multi-core commodity CPUs are outpacing what it can do with its in-house FPGA hardware. The second wind is blowing more and more unstructured data into the worlds of high-performance computing and also into media processing and the web applications area.
DataDirect has responded to this with FPGA-based S2A controller architecture storage by making IBM's General Parallel File System available with its GridScaler product and the Lustre File System available with its ExaScalar product.
The S2A architecture has a field-programmable Gate array (FPGA) controller looking after StorageScalar 6000 drive enclosures. However, Nehalem-based network-attached storage (NAS) storage arrays and file virtualisation switches are appearing which have better random I/O performance than DataDirect's S2A systems.
In June DataDirect announced a switch to a new controller design called its Storage Fusion Architecture. It is now shipping the first product using it, the SFA 10000E Open Storage Appliance, nicely and numerically capping the preceding S2A 9000 line of products.
The S2A9900 delivers 6GB/sec of transactional I/O. The SFA 10000E increases that to 10GB/sec but can also deliver more than one million IOPS of random I/O performance.
It does this with a pair of Nehalem processors, x86 servers really as the SFA controllers use gen 2 PCIe busses and DDR3 RAM. As with the S2A 9000 controller, up to 1,200 disk devices are supported, meaning SATA and SAS hard drives and solid state drives.
The latest 2TB SATA drives are supported, giving a maximum capacity of 2.4PB per controller. The one million IOPS are to/from a controller cache with backend disk IOPS being around 300,000.
The Nehalem cores and DataDirect's threaded code means that the SFA controller can do more than the S2A controller. Data Direct says it will offer SFA 1000E implementations of both the ExaScaler and GridScaler platforms. They will be offered with either InfiniBand or 10Gb Ethernet LAN connectivity options to deliver over 5GB/sec of performance per SFA10000E building block. SFA 1000Es can be aggregated within a clustered storage environment to deliver more than 200GB/sec.
Jeff Denworth, DataDirect's marketing veep, says: "By embedding these popular HPC file systems natively into this scale-out platform, we achieve a dramatic reduction in storage cluster components, I/O latency and most importantly - new levels of cost savings and management efficiency for our customers.”
DataDirect is entering a pretty crowded field. Isilon sells clustered NAS nodes with its latest, the IQ 72000X offering 144-node clusters and a 10PB single volume with aggregate bandwidth passing 40GB/sec. The NAS-specific IQ 72000S offers, Isilon says, over 1.7 million IOPS with aggregate throughput of more than 45GB/sec. Exanet also offers a scale-out NAS cluster and StorSpeed has an SP5000 clustered file caching device offering 350,000 IOPS.
BlueArc is another high-performance NAS system supplier as is Avere with its NAS access accelerator product. Then there are Symantec's FileStore product and NetApp with its cluster-capable ONTAP 8 release.
If that's not enough, let's add IBM and its XIV-based scale-out file service and HP's acquisition of Ibrix and PolyServe. That's nine suppliers, from tier one storage vendors, though established niche players, to start-ups all aiming to offer the most scale-out file-serving bangs for the lowest bucks.
Every supplier sells on bangs per buck; performance is king in this area. Raw IOPS and bandwidth figures are probably not the best means of comparing these systems.
Avere, BlueArc, Exanet, Isilon, NetApp and Symantec use the SPEC sfs2008 benchmark and that provides a cost, performance and configuration comparison against a standard NFS scenario. It will be good if DataDirect submits its new SFA 10000E product to this benchmark as well. ®