Product and service reviews are conducted independently by our editorial team, but we sometimes make money when you click on links. Learn more.

DDN's Flashscale Hyperconverged AFA Targets Big Data Analytics, HPC

By - Source: Toms IT Pro
Tags :

DDN may be the biggest storage company of which you have not heard. It's the world's largest private storage company. DDN's devices are used by many global enterprises, including J.P. Morgan, Boeing, Suzuki, and so on; the company also does backup for some of Apple's cloud data, and support for some unmentioned large financial institution's high frequency trading programs. 

Available in August of this year, the DDN Flashscale Hyperconverged All-Flash Array delivers up to 6 million IOPS with as much as 60 GB/s throughput in 4U for as low as $1/GB. This platform is designed for enterprise big data and analytics, financial services, web-scale cloud, large databases and high performance computing (HPC) environments.

What makes DDN's Flashscale so special?

  • It offers low latency (on the order of 100μs)
  • It uses Intel Broadwell Procesors, and the fabric is based on PCIe 3
  • It has the ability to directly execute applications
  • A 4U box may contain up to 72 SAS or 48 NVMe SSDs
  • Connectivity options include Infiniband, Fibre Channel 16/32, Gb/E 10/40/, or OmniPath
  • Each 4U adds 576 TB capacity, natively and additional units may be added to fill a rack (to 7.3 PB, 60 M IOPS, 600 GB/s throughput) or even multiple racks (17 racks yield 1 B IOPS, 10 TB/s total throughput, 46 PB storage)

We spoke with Molly Rector, CMO of DDN, who was unabashedly enthusiastic about Flashscale's presumptive role in the new hyperconverged world. Hyperconvergence is an infrastructure that tightly integrates compute, storage, networking and virtualization resources via software and APIs, typically in a common hardware box or even extending to a network. Rector focused on several key points.

The fabric and design of the motherboard permits very high density of flash drives to fit into a 4U space, and the redundant fabric design provides higher throughput than Intel's limit, presumably by running multiple channels simultaneously. By putting common drives in a very dense, efficient package, Flashscale combines efficient IOPS with best cost-per-capacity and cost-performance. Large storage users, especially those with high-scale performance needs make use of Flashscale for those reasons and because it is possible to run software on the storage device. Running embedded software on top of the storage device allows interrupt free, in-storage, virtualized computing. The burst buffer and data aware cache also provide memory and application acceleration, sufficient for high frequency trading, genomic pattern searching, and HPC.

Intel's Broadwell application processors running on the PCIe 3 embedded fabric are able to run Hadoop, OpenStack and multiple file systems. Infinite Memory Engine (IME) is DDN's proprietary software that virtualizes NVMe SSDs into a single pool of shared memory. IME sits between compute nodes and the parallel file system (PFS). This is what enables I/O, PFS and application acceleration.

IME does three things:

  • The write accelerating burst buffer "absorbs" data into the solid state cache faster than a file system could absorb it
  • IME reorders application I/O to optimize cache flushing to long term storage, thus optimizing applications and accelerating the file system
  • Out-of-band API configuration of the IME appliance optimizes both reads and writes, which not only shortens the job queue, but also permits more simultaneous job runs, which shortens user run time

Some of the examples given by DDN that make good use of Flashscale include massive data analytic projects, such as searching for mobile devices (over 500 billion), a global Content Delivery Network (CDN) used by a cable provider, nation-wide cloud computing for a very large Japanese company, and an interesting analytic use case of SAS software, where 400 percent better utilization of the analytics package on faster customer workflows led to significant software use cost savings.

Is DDN's new AFA a game changer? For enterprises, this combination of throughput, cost saving, and fast storage is evolutionary, not revolutionary. More enterprises are adding flash storage to their critical systems, but not yet converting all storage to flash. 

More information on Flashscale can be found at