05 Jun Intel Is Disrupting The Storage Tier

The traditional data tier is changing and I’ve been talking about this for a while. I’ve written about the explosion of data,  and how to craft a strategy to exploit that explosion of data.  That defines how to talk about data, but what about the storage of data? Intel announced two new product families last week that  I believe will disrupt the physical storage tier. These announcements open the door to new possibilities in managing an organization’s explosion of data.

Intel first announced the coming availability of Optane DC Persistent Memory.  Optane DC Persistent Memory brings very high-speed and low-latency non-volatile memory to a system’s memory bus.  Think of this as an SSD that lives in a DDR4 DIMM slot and is written to as if it is system memory.

At the other end of the spectrum, Intel disclosed their progress with QLC NAND technology and offered a glimpse of products that are expected to arrive in the second half of 2018.  QLC is a new type of NAND technology that allows much higher-density and cost-efficient SSDs that can be delivered from the more dominant TLC NAND-based SSDs. The benefits of QLC come with trade-offs which limit their primary utility to read-intensive workloads.  Read-intensive workloads are precisely where hard disk drives tend to live today.

The power of Persistent Memory

Intel jump-started the conversation around performance storage and persistent memory in 2015 when it, along with Micron Technology, announced a new type of byte addressable non-volatile memory called 3D XPOINT. This technology promised very high-speed, low-latency persistence that can live not just in traditional SSDs, but also in a system’s high-speed memory bus.

Persistent memory, which is sometimes referred to as Server Class Memory (SCM), isn’t treated by software as if it is a disk drive, but rather as memory that can survive a reboot.  The benefits of this are immense.

At the launch event for Optane DC Persistent Memory, Intel demonstrated a NoSQL in-memory database from Aerospike that used persistent memory to cache state between server restarts.  The traditional scenario, using DRAM and SSDs, yielded a restart time of 35 minutes. Using Optane DC Persistent Memory to cache the persistent state allowed the database to restart in an astonishing 17 seconds.

The power of persistent memory is evident to the industry. The Storage Networking Industry Association, or SNIA, has long had a working group focused on defining software interfaces.  The ISV and OSV communities are responding rapidly, with both Microsoft and the Linux Community adopting and integrating Intel’s interface libraries into their offerings.

Microsoft began supporting the persistent memory in Windows Server in late 2017 and is today shipping test builds of the next version of Windows Server with more sophisticated support. Microsoft also supports persistent memory in SQL Server 2016 SP1, and in its Storage Spaces Direct feature for HCI and software-defined storage enablement.

The broader ISV community also sees the tremendous benefit inherent in the approach.  Today Oracle, Aerospike, SAP, Cassandra, and RocksDB all support persistent memory models.

Intel will be first to market with DDR4 persistent memory, setting the bar for those that follow.

This is a new tier zero for data access, and that is already changing how the software world views storage.   Intel is sampling parts now to its OEM partners and expects to ship for revenue later this year.

QLC NAND density

Fast storage gets all of the attention, but the real workhorses in the data center are mid-tier storage arrays that continuously serve data into the enterprise.  These arrays don’t need the performance, or the cost inefficiencies, offered by SSDs built on today’s TLC NAND technology. These devices are the home of spinning hard disk drives.  SSDs are overkill for the application. That’s changing.

Intel and Micron Technology have been working together on a new NAND technology called QLC, and the two companies jointly announced the qualification of the first parts in late May.  It’s not important what these acronyms stand for, or what the underlying physics of these devices entail.  What is essential is what QLC NAND brings to the table for SSDs deployed into the data center.

QLC NAND provides high-density storage, approximately 33% more density than is offered by TLC NAND. The industry expects SSDs to arrive with capacities between 2-8TB in a 2.5″ drive form-factor.  Because of the efficiencies of the process, this increased density does not come at a premium. QLC offers higher density at a lower cost per bit than TLC NAND.

While this high-density/low-cost dynamic may seem like a massive win for SSDs across the board, there are limitations to the QLC NAND that make it unattractive in a high-performance SSD.  QLC NAND has reduced write endurance. The technology is also performance limited in certain workloads. Random writes, for example, are a performance deficit as compared to TLC NAND.

What this means, practically speaking, is that SSDs built with QLC NAND technology are an ideal fit for HDD replacement in mid-tier storage arrays.  The performance and density are better than are offered by hard disk drives, and the economics start to make much sense. Hard disk drives become lower-tier offerings as the world moves to solid-state devices.

Intel’s execution

Intel’s non-volatile memory solutions group is executing to a vision that disrupts the traditional storage tier.

Intel is tying persistent memory technology to upcoming Xeon parts that will deliver the benefits of the technology in a way that providers who do not have a full stack of hardware will be challenged to match.  Intel’s approach to persistent memory redefines tier zero storage. Persistence in a DDR4 DIMM slot is a massive win for in-memory computing.

Below this, Intel continues to evolve Optane SSDs, which provide very high-performance products based on 3D XPOINT technology.  Optane SSDs supplement traditional TLC NAND-based SSDs by providing a high-performance where it is needed. Optane doesn’t yet have the density or economics of a more traditional 3D NAND based SSD, but it serves a critical need in performance systems.

Intel continues to evolve TLC NAND.  Lost in the flurry of announcements this month was Intel’s news of a process improvement for TLC which serves to increase both density and overall performance. These process improvements will arrive in products later this year.

At the other end of the spectrum, Intel is on the verge of disrupting the hard disk drive market with its upcoming QLC NAND based flash products.  These parts will fit into mid-tier storage arrays, and redefine the performance characteristics of that market.

Intel isn’t the only storage vendor innovating.  Much of Intel’s NAND and 3D XPOINT process technology is being co-developed with Micron Technology, which is releasing products on its cadence.  What is remarkable about Intel’s efforts is the wide breadth of solutions, along with the natural integration with its efforts in evolving server architecture.  There’s value in delivering a full-stack solution, and Intel is leveraging that value.

Note: This blog contains significant contributions by Moor Insights & Strategy senior storage analyst Steven McDowell.