Flash Solutions Taking the Data Center by Storm

By Michael Vizard
flash memory market

Flash solutions offer a study in contrasts. The flash memory market is rife with opportunities but also complexity and challenges.

As flash solutions have taken the data center by storm, most IT organizations have taken a "flash-first" approach to data storage. Assuming the price is right and the workload they need to deploy can fit within the amount of capacity available on a flash storage system, the default option for primary storage these days is now some form of flash memory.

Flash Solutions: Obstacles Amid Opportunities

While selling flash storage these days is relatively straightforward, the challenge facing solution providers and their customers today is that there is no shortage of options spanning multiple types of solid-state drives (SSDs) and flash memory modules.

With the arrival of flash storage systems capable of holding 60TB per drive, the rate at which flash solutions are replacing traditional magnetic storage is only going to accelerate in the years ahead, especially once these devices can store as much as 100TB per SSD.

However, besides the order of magnitude in application performance provided by flash storage systems, the major benefits enabled by the shift to flash memory include a dramatic consolidation in the amount of physical space consumed in any given data center. That space reduction, in turn, reduces the number of square feet that needs to be acquired as well as the amount of power required for flash storage.

In addition, the need for expertise associated with optimizing I/O performance on spinning magnetic storage devices is eliminated. Instead of wasting massive amounts of storage capacity to make sure any given application has optimal access to where data is placed on a magnetic disk drive, all the data stored on a flash memory device is equally accessible.

Flash storage eliminates the need for the post-processing of data in a way that simplifies data management, explained Lee Caswell, vice president of products for the storage and availability unit of VMware. "Customer satisfaction winds up being higher while support costs are lower," Caswell said.

Couple those ancillary benefits with dramatic improvements in application performance, and the shift to flash solutions is a no-brainer. A new report from Trendfocus finds that the enterprise SSD market will reach 4.068 million units for the second quarter of 2016, an increase of 10 percent quarter-over-quarter. In total, Trendfocus estimates that 4.16 exabytes of flash storage were shipped in the quarter.

Things start to get more complicated when trying figuring out where to optimally place flash storage based on the requirements of different types of application workloads.

The flash market and technology continue to evolve. Initially, flash storage showed up on servers. Applications and databases running on a single server benefited immensely from higher levels of direct-attached flash storage.

These days, flash memory is also made available on storage arrays. Depending on the application workload and the amount of available IT budget, flash storage manifests itself either as a layer of cache on top of traditional magnetic disk storage or in the form of an all-flash array. The benefit of an array is that flash storage becomes accessible to applications running across multiple servers.

To make those storage arrays even more appealing, it's now possible to plug flash storage arrays into NVMe interfaces that provide the throughput needed to keep pace with flash storage technologies. Meanwhile, that same NVMe interface is also being used by vendors, such as Tegile Systems, to drive hyperconvergence into a traditional rack system in the data center.

"We're able to scale out using up to eight controllers," said Rob Commins, vice president of marketing for Tegile Systems. "We can get up to 6 petabytes of raw capacity."

Flash Solution Delivery Options

One challenge flash solution providers face is that they need to figure out whether it's better to deliver flash storage as an SSD or a module. An SSD essentially presents flash storage as a much faster disk drive to an application. A flash module dispenses with that legacy disk architecture to provide much higher levels of I/O performance.

For example, IBM uses flash modules to deliver an array for the midmarket priced below $1.50 per gigabyte. More recently, IBM unveiled an all-flash array aimed at hyperscale environments.

"We think the time for the all-flash data center is here," said Alex Chen, director of storage systems and offering executive for file and object storage at IBM.

Sorting out different price-performance levels associated with various classes of SSDs and flash memory modules requires a fair amount of expertise on solution providers' part. Most IT organizations are still sensitive to the cost per gigabyte of storage acquired.

"I would say that price per gigabyte is most often the most limiting factor when moving to flash," said Shawn Scanlon, director of solutions architects for HPM Networks. "But at least 80 percent of the time now customers want to be able to use some kind of flash."

However, while flash storage today is roughly equivalent to a magnetic disk in terms of cost per gigabyte, the sticker shock associated with replacing thousands of gigabytes of magnetic disk storage can be considerable. Because of that, solution provider HPM Networks has found that all-flash arrays from Exablox that allow IT organizations to employ third-party SSDs at price points in the range of 50 cents per gigabyte is an attractive offering, Scanlon said. In contrast, other vendors tend to mark up the actual cost of the SSD by 100 percent or more, making the overall solution prohibitively expensive, he said.

Other solution providers, however, steer clear of the cost per gigabyte discussion.Consiliant Technologies CEO Dave Cerniglia said cost per gigabyte is a legacy mindset left over from when hard disks dominated enterprise storage.

"From a customer perspective, it's noise," Cerniglia said. "It winds up being a complicated question because of how the vendor twists maintenance fees, data deduplication and compression."

Because of these issues, Consiliant Technologies chose to partner with Kaminario for all-flash arrays because Kaminario guarantees I/O performance levels, Cerniglia said.

A Flash Memory Market for SMBs

In general, there is now a concerted effort to expand the appeal of flash technology into the small- and midsize-business (SMB) market. For example, Hewlett Packard Enterprise (HPE) has already pushed the cost of its flash storage arrays to under $10,000 as part of an effort to increase the appeal of the technology to SMBs—a market segment where the channel tends to wield the most influence.

"We think this will create a major new opportunity for our channel partners," said Brad Parks, director of go-to-market strategy and enablement for HPE Storage.

Flash storage is only the beginning of a major transformation from traditional memory architectures. The cost of dynamic RAMs continues to drop, while new forms of non-volatile memory such as the 3D X Point architecture are being built by Intel and Micron that promise to accelerate the rise of in-memory computing altogether.

It may take time before those technologies go mainstream, but it's already apparent that traditional magnetic storage devices are being relegated to secondary and tertiary tiers of storage in new deployments. That doesn't mean, however, that magnetic storage systems will disappear from the IT landscape overnight.

"Flash is in the process of replacing hard disks in tier-one storage," said David Hill, principal analyst with the Mesabi Group. "But that doesn't mean that every tier-one application will need flash."

In the meantime, solution providers across the channel that are looking to tap the most lucrative segments of the flash memory market would be well advised to put themselves in a position to proactively drive the shift to in-memory computing in all its forms today versus waiting for IT organizations to embrace flash memory one new application workload at a time.

Mike Vizard has covered IT for more than 25 years, and has edited or contributed to a number of tech publications, including InfoWorld, CRN and eWEEK. He currently blogs daily for IT Business Edge and contributes to CIOinsight, Channel Insider and Baseline.

This article was originally published on 2016-08-31