Legacy storage vendors have faced a unique challenge as they tried to address the emerging flash opportunity. Unlike startups that could essentially start with a “blank sheet of paper”, these vendors had existing customers to care for. Their first mission was to try to deliver flash performance to their installed customer base, a strategy that led to a slower, “evolutionary” flash roll-out that didn’t always meet customer demands.
Companies like NetApp started their flash journey by offering drive form-factor, solid state disk drives (SSD) as an independent tier of storage. NetApp was also one of the first legacy vendors to offer a caching solution, in the form of their Performance Acceleration Module (PAM). This card was a PCIe based flash solution that provided a second layer cache to hold blocks evicted from the WAFL buffer cache.
NetApp FAS Hybrid Arrays enable users to create a volume flash SSDs and allowed data to automatically move in and out of that tier. Eventually, NetApp introduced the FAS All-Flash which can be deployed as a stand-alone system or as a high performance tier in a clustered ONTAP configuration. Like NetApp Hybrid Arrays, it provides the ability to non-disruptively move data between tiers, including the cloud.
At the same time NetApp’s other product line, the EF-Series, also began to integrate flash and eventually delivered an all-flash option. The EF series is focused on extreme high performance and high reliability. But does so without features but massive performance, about 450k IOPS per system. If you’re counting, that’s four flash products in the NetApp portfolio. All of these solutions were designed to extend the current NetApp operating system capabilities to include flash and to allow their customers to continue to leverage the NetApp data management tools they have come to rely on.
Unfortunately with flash any amount of overhead causes latency, latency can lead to a performance problem in the flash storage tier. For many data centers the capabilities provided by NetApp’s data management software far outweigh the potential performance problems this creates. But there are those where any latency can be problematic and this is where the EF series fits in. But NetApp felt there was room for middle ground, a performance focused flash array with in-line data efficiency.
To address this market, NetApp recently introduced the FlashRay, a purpose built all-flash storage solution designed from the ground up to support memory based storage. This introduction also includes a new operating system which NetApp has dubbed “Mars”. NetApp claims the system provides single node performance of 250K IOPS and provides inline data efficiency (deduplication and compression).
While in its initial targeted release, the system is available as a single node only. NetApp claims that it will be providing scale-out capabilities in the near future as well as integrated data migration between a ONTAP-based FAS and a Mars-based FlashRay. This would allow NetApp customers to seamlessly move workloads between full featured FAS based systems and highly performing and efficient FlashRay based systems as needed.
Does NetApp have a strategy or is it just throwing products on the market to see what will stick? For the larger storage vendors, having multiple product offerings is not unusual and most of the major players now have a dedicated, purpose built all-flash solution. For these vendors, a multi-pronged flash strategy is required. They have invested decades in their software tools that provide data management, and extending those tools to include flash support makes sense. But they’re all feeling the pressure from the startup flash vendors and there are times where that legacy, feature-rich storage operating system may get in the way of full flash performance.
The multi-pronged approach gives customers choice. They can choose features with solid performance, they can choose maximum performance with minimal features or they can choose data efficient flash. Most workloads will just find the feature rich situation, but giving customers a choice is the ideal answer, especially if it includes the ability to transparently move data between the two solutions.