raywoo - Fotolia

News Stay informed about the latest enterprise technology news and product updates.

DataDirect Networks combines IBM GPFS, Storage Fusion for HPC

DataDirect Networks packages IBM General Parallel File System with its Storage Fusion Architecture for scale-out NAS storage.

High-performance computing specialist DataDirect Networks today ushered to market a parallel file system appliance aimed at scale-out storage for big data enterprises.

DataDirect Networks (DDN) said its branded GS7K Parallel File System storage device will be generally available in December and target broadcast media, financial services, government, life sciences, oil and gas, and other high-performance computing (HPC) sectors.

The GS7K design embeds IBM General Parallel File System (GPFS) in DDN's Storage Fusion Architecture (SFA) operating system.

The base GS7K enclosure houses 60 drives in a 4U rack space. Users can scale storage linearly by adding up to four 84-drive nodes for a maximum of 20U racks and 396 drives, including a mix of solid-state drives (SSDs), SATA, SAS and near-line SAS disk. Each 84-drive rack provides 504 TB of raw capacity with rated throughput of 12 gigabits per second (Gbps).

The design aims to tackle point-to-point bottlenecks of NAS systems caused by external servers and network components. Client machines in NAS systems are limited to communicating with a single server at a time. DDN's appliance runs file-serving software inside the storage controller. The GS7K system stripes data across multiple systems in a cluster so reads occur in parallel.

Protocol support is provided for Network File System, Common Internet File System and Server Message Block shares over four 56 Gbps InfiniBand ports.

Storage density, performance drive scale-out architecture

The SFA box devotes approximately one-third of CPU processing power to manage RAID data protection and uses the remaining processing to host virtual machines in the open source Kernel-based Machine (KVM) hypervisor. IBM GPFS uses KVM to talk directly to storage.

"Our sweet spot is the class of big data customers that need to scale the performance of NAS," said Uday Mohan, DDN's senior product marketing manager. "Customers told us they want the performance of a parallel file system in a scale-out box that has data management features they're used to, yet is simple to manage and install."

Mohan said the GS7K is engineered for density and performance to exploit the scalability and data management features inherent in GPFS, including quota management, snapshots, rollback, and synchronous replication of data and metadata.

"There are a whole bunch of enterprise data management features that customers are used to from NAS that we had to build into this product," Mohan said.

The file appliance is a natural extension of DDN's focus on HPC, said Terri McClure, a senior analyst at Enterprise Strategy Group (ESG).

"The markets they're going after are all about saving significant money and making significant revenue if I can get my results quicker -- that's an HPC-like message," McClure said.

"To me, the challenge is going to be proving it in the market," she added. "GPFS is both a blessing and a curse. It's incredibly powerful and flexible with high performance, but it's also very hard to use. When people think of GPFS, the first thing they think of is complexity."

SFA fuses IBM GPFS with WOS for the cloud

Mohan said DDN made several optimizations to its SFA to augment the native data management of GPFS. It unveiled an API for using the DDN Web Object Scaler (WOS) platform as a cloud storage tier for asynchronous, over-the-wire replication.

The API turns WOS into a wide-area distribution mechanism for sharing data in a follow-the-sun model for geographically distributed work teams. Users are able to federate up to eight GS7K clusters in a single namespace and move data between SSDs, rotating media, tape and the WOS cloud.

"The bridge technology means you can use the DDN solution on the edge for really fast scratch space and migrate your longer-term data off the WOS and not have to use scratch space for data," ESG's McClure said. "That gives you an opportunity to re-use data and have it centralized for multiple groups."

The GS7K integrates DDN's Real-time Adaptive Cache Technology (ReACT) for real-time workload analysis that optimizes cache by managing the ratio of small files to large files. ReACT bypasses large sequential files directly to spindled media and reserves fast cache for applications with small randomized I/O. Additionally, the Storage Fusion Xcelerator tool in SFA accelerates block and file-based data access by pre-populating data from hard disk drives to SSDs in cache.

DDN's quality of service engine guarantees a dedicated response time to storage of approximately 80 milliseconds even if a drive is unresponsive. The system uses parity drives to build data if other drives fail.

DDN has not made a price list available for the GS7K, but Mohan said the company is aiming to deliver storage at a cost of $250 per TB.

Next Steps

DDN adds global erasure coding to WOS

Web Object Scaler adds smart search feature

Dig Deeper on Data storage strategy