All-flash arrays are moving from niche storage performance systems focused on revenue-generating applications to mainstream storage systems that handle a wide variety of workloads.
But all-flash array vendors need to provide not only essential features, like snapshots and thin provisioning, but enterprise features, such as active-active controllers and replication, to make all-flash data centers a reality. They also need to add AFA-specific characteristics such as quality of service (QoS), performance analytics and integration to other data storage platforms.
Phase one: The basics
When all-flash arrays (AFAs) were introduced two years ago, the primary focus of vendors was cost per gigabyte. Thanks to the decreasing cost of NAND flash and data efficiency technologies like deduplication and compression, almost every all-flash array vendor was able to make its products more price-competitive than performance-oriented, hard disk-only systems. Data efficiency and the continual decrease in flash prices make the effective cost per gigabyte of all-flash arrays suitable for both high-performance and mainstream workloads.
When initially released, most systems only provided the above data efficiency features, as well as thin provisioning and snapshots. But mainstream workloads require AFAs to provide more complete features to bring arrays on par with legacy enterprise storage systems.
Phase two: Enterprise parity
Enterprises expect high-availability features, such as active-active controllers and replication, because the applications and workloads that justify AFA performance are almost always mission-critical. Keeping these applications up and running is equally important, if not more so, than providing them with the performance they need. Today, the majority of all-flash arrays provide basic data protection from the failure of a single SSD, and many provide protection from a controller failure. But enterprises need to pay attention to how all-flash array vendors deliver this protection.
To provide controller availability, some AFAs use an active-passive controller setup. This directs all workloads at a single controller, and the second controller is not used unless the first one fails. The advantage of this technique is applications receive the same performance in a failed state as they would in an active state.
The alternative is an active-active design. In this configuration, volumes or LUNs are assigned between controllers. During normal operating conditions, both controllers deliver performance. If there is a failure, the surviving controller is assigned to the other controller's LUNs and existing assignments. There are two disadvantages to this setup:
- The surviving controller has to do double duty during a failure, so application performance could suffer.
- The IT administrator has to manually balance workloads between the controllers so neither one is overwhelmed.
The other aspect of high availability is replication for disaster recovery, but several all-flash array vendors are still missing this feature. The lack of a built-in ability to create a DR copy is a concern for many enterprises, but there may be some advantages to not having this feature. If the enterprise leverages a third-party software platform instead of a built-in product, it typically gains better OS and application support, as well as the ability to replicate from an AFA to an HDD-based system at a DR site.
Phase three: Advanced features
IT planners should expect more from their all-flash array vendors than just enterprise feature parity. AFAs should have the following advanced features that let IT planners more intelligently manage their investment and derive a greater ROI:
- Analytics. Depending on the depth of the reporting, analytics can provide the insight IT professionals need to ensure their system delivers maximum benefits.
- Quality of service. QoS can take analytics information and build safeguards to ensure mission-critical applications always get the performance they need. Because IT admins buy AFAs to address specific mission-critical application performance problems, QoS allows more mainstream applications to benefit from flash performance without affecting the core reason for the purchase.
- Integration with other platforms. Studies indicate that as much as 80% of a company's data has not been accessed in more than a year. AFAs need to provide the capability to move this data to secondary storage systems. To that end, some AFAs are integrating with object storage systems, so older data, snapshots and backups can be pushed to cost-effective, highly scalable object storage systems.
- Integration of 3D triple-level cell (TLC) NAND into the array by the vendor. While more affordable than multi-level cell (MLC), 3D TLC may be less durable. Some vendors are overcompensating for this by providing excess flash capacity that is kept in reserve. The extra, unallocated capacity allows for wider distribution of writes, which extends the durability of the system. Other vendors are placing an MLC tier in front of the 3D TLC tier. In this scenario, the MLC tier acts as a shock absorber for the 3D TLC tier. IT planners should verify that their AFA vendors have a plan for 3D TLC and are comfortable with how they will mitigate the technology's weaknesses.
- Scale-out all-flash arrays. These arrays can expand their performance and capacity by adding nodes. The value of these products is largely dependent on the data center's growth trajectory. Many organizations find that a single scale-up AFA will meet performance and capacity demands for years to come. If an organization expects to grow quickly and sporadically, it should consider a scale-out system.
All-flash array vendors have come a long way, evolving from point-performance products to traditional data center offerings. In some cases, vendors are making an argument that AFAs are good for archive data and active data. As IT planners consider the broader adoption of all-flash arrays, they should look for capabilities that not only bring these arrays on par with mainstream HDD systems, but provide advanced features that enable organizations to extract the maximum value out of their AFA investments.
The big vendors to look out for in the all-flash array market
Storage and all-flash arrays: A big-picture approach is best
AFAs could become the only game in town for primary storage