Cloud and Enterprise SSDs – Where is 2020 taking us?

December 16, 2019

         I think everyone in the storage industry agrees, more needs to be done to enable the world to take full advantage of Solid State Drives (SSDs).  SSDs have gained significant traction in providing the performance improvement needed in storage devices that Hard Disk Drives (HDDs) have failed to deliver.   To encourage adoption, the interfaces and design of SSDs has tracked closely with HDD system requirements.  Similarly, the “one size fits all” approach to the HDD market has spilled over into the SSD market at a time when cloud based and purpose-built systems cry for a different approach.  This has led to the point where many have begun building their own SSDs (eliminating a manufacturer’s margin and matching capabilities to their needs).

              2020 appears to be a year in which significant advances in addressing the uniqueness of solid state technology and realizing its capabilities will be made.  We can be confident of that as we see the following movements taking place in the market:

1. The trend to building your own SSDs is accelerating.

             In addition to the product attributes SSDs seemed to have inherited from HDDs, the business models seem to have stuck as well.  That is, the HDD manufacturers embraced the vertical integration model necessary to the low cost provider of HDDs.  Media, controller ASICs and firmware, read and write heads, etc., are largely developed and built by the HDD manufacturer themselves.  With the sensitivity, complexity, and capital cost of building a precision electro-mechanical device like and HDD, only the HDD provider (and we’re down to 3 worldwide) could build the finished hard drive.

               SSDs do not have the manufacturing complexity of hard drives.   In fact, many of the contract manufacturers building PC, server, and other circuit boards can build SSDs.  Numerous companies are selling SSD controller chips.  Robust firmware is also needed but several of the cloud hyperscalers are writing firmware/software to suit their needs and Burlywood is making its mark filling this need.  The question of why buy anything other than media (Flash) from the current vertically integrated SSD suppliers is being asked more and more

2. The need to Integrate/reconcile system software is being addressed.

Speed improvements, the new requirements of solid state media (Flash), and the proliferation of new applications and use cases for storage have all provided cause for a new look at the system/storage device interaction.  Work being done at the system (host side) level might be better handled closer to the storage device and vice versa.  Redundancies are often present and can be reduced or eliminated.  Western Digital, in a blog post notes:

“With the advent of cloud storage and the proliferation of analytics, edge devices and machine learning, IT organizations have to optimize their applications and infrastructure for new types of workloads” 

Similarly Micron Technology owns up to the opportunity in their own post saying:

“As hardware continues to progress in leaps and bounds, getting full utilization of your hardware with legacy software can be a challenge.”

         The good news we see in 2020 is recognition that there is opportunity as system software and workloads get reconciled with SSD device software.  Performance, cost, and endurance of the SSD media will all improve as these advances are made.  Evidence of the progress can be seen in the early stages of the work done with the NVMe protocol and in-storage computing.   The release of NVMe 1.4, for example, includes a cornucopia of new, optional features to enable the system to alter the SSD operational mode to match.  Implementation of these features is difficult, as there is currently no one place in the system where the knowledge exists to tell the SSD how to behave. Burlywood, however, is delivering much of this integration and reconciliation today in our storage device (SSD controller) software.

3. Increasing awareness of the need to monitor and predict failures in SSDs.

While the current crop of large SSD manufacturers may not be required to build everyone’s SSD, they are key to the continued development of solid state media.   They continue to increase 3D flash die layers and add bits to each storage cell.  The positive effect of this work allows solid state technology to continue its trek to lower and lower costs.   The potential downside of this improvement in storage density is the reduced usable life of the flash media.  In a paper published by Intel, they warn of SSD unique failure repercussions.

“… with the industry shift toward the faster storage medium of solid state drives (SSDs), the very nature of the threat of disk failure is changing. Data from failed SSDs cannot be recovered, as it can often be with harddisk drives (HDDs). Organizations that are adopting SSD storage must therefore compensate for this difference with additional precautionary measures just to retain the same level of risk for data loss as before”

With the growing awareness of flash media life considerations and potential failure impacts, storage engineers are beginning to appreciate the monitoring of device characteristics that will help predict failure.  Anticipation of a reduced life of the device or flash media itself could also allow for redeployment delivering longer life (and better economics).  Burlywood is  the early leader in delivering the enhanced telemetry key to enabling these SSD analytics.

With all this going on, 2020 will be a good one in delivering against the promise of solid state technology.