TORONTO – A vendor that normally provides flash for enterprise storage systems is taking on data centers directly with a new platform designed to address an emerging market that research firm IDC has dubbed “big data flash.”
SanDisk recently announced its InfiniFlash storage system, which aims to provide flash-based capacity for big data and hyperscale workloads at a price point competitive with spinning disk. Rather than use SSDs, InfiniFlash uses 55 hot swappable cards, each with 8TB of flash capacity, to provide 512TB of raw flash storage in a 3U enclosure. The unit can connect with as many as eight off-the-shelf servers using SAS connections.
At a live web cast for the product launch, Eric Burgener, research director of IDC storage practice, said there is a large market opportunity that demands a category of storage that does not exist. The characteristics of big data flash storage include performance consistently better than hard drives at scale with hundreds of petabytes; under a $1 per gigabyte; and, implementations that meet standard enterprise requirements such as scalability and availability. He said target applications include big data environments, including analytics, as well as content repositories and media streaming.
In an interview with EE Times, Gary Lyng, SanDisk’s senior director for product marketing, said the design of InfiniFlash was driven by input from users such as Web 2.0 customers and ultimately focused on being simple while addressing the cost of flash per gigabyte. Another key design requirement was making it hot swappable and redundant.
SanDisk has incorporates some of its existing technologies into the InfiniFlash platform, including those that have been acquired in recent years. It includes SanDisk’s ION Accelerator software stack from Fusion-io for block storage applications. For scale-out block and object storage workloads, it uses the open source CEPH platform to provide enterprise-class data services. It also includes development libraries and a software development kit that allow customers to optimize applications for use with the system.
At launch, there are three versions of the platform: the IF100, IF500 and IF700. They vary in terms of what software and services are available on top of the hardware. The IF100 targets OEMs, integrators and resellers that want to include the storage array as part of their own offerings, while the IF500 is intended for scale-out storage applications, and includes an optimized version of CEPH. The IF700 integrates Ion Accelerator technology and targets high-performance applications.
Roark Hiloman, engineering fellow with SanDisk’s systems and software solutions group, said the individual flash chips are designed to be used in aggregation to get the performance benefits. The primary design challenge was achieving the density. “A lot of work went into the firmware to get 8TB.”
Richard Fichera, VP and principal analyst for infrastructure and operations at Forrester Research said SanDisk is essentially going out and competing with its own customers – what he called a “pretty ballsy move.” It’s a strategic and smart move for the company, and brings flash into a wider footprint of use cases, while accelerating the price competitiveness of flash at approximate $1 per raw gigabyte.
Fichera said SanDisk InfiniFlash shares characteristics and levels of performance with all-flash arrays in that it does not have to be concerned about what data is hot and provides the same speed of access to all data on the array, rather than caching what’s important on flash as a hybrid flash array does. It’s also packing memory in circuit boards as efficiently as is possible, he said, much more so than can be done with SSDs.
Although proprietary flash architectures have their unique characteristics, said Fichera, they are not necessarily more complex to manage than a storage array with SSDs, which have their own nuances depending on the vendor. Enterprises also tend to make an investment over the lifetime of a technology, and while InfiniFlash is new, different and proprietary, there’s little risk of SanDisk not supporting it in the long term, he said, adding that enterprises have been willing to take risks with startups.
SSDs have always been a “botched” use of flash, said Jim Handy, principal analyst with Objective Analysis, in that interface sandbags what the flash is capable of by applying a hard drive architecture. “The interfaces are designed for how hard drives like to operate.”
Using raw flash rather than SSDs for enterprise storage systems isn’t new. Others such as Violin, Skyera (now part of HGST) and IBM have developed arrays that incorporate flash chips directly, but this is new territory for SanDisk, said Handy. “The biggest issue facing SanDisk is it is not already in this business. It will have to establish sales channels and brand recognition.”