Trends and Driving Forces
By Rainer W. Kaese, senior manager business development, Toshiba Electronics Europe, storage products division, who has been with Toshiba for over 20 years, initially specialized in application specific ICs, managing the ASIC design center, and later the business development team for ASIC and foundry products.
There has been much written about the demise of the HDD, with nearly all new mobile devices using flash memory as the main storage medium.
The adoption of flash-based SSDs in laptops and desktop computers continues at pace and also continues to eat into traditional HDD applications.
Despite this shift, it is expected that HDDs will still account for the majority of sales in 2016, as HDDs remain the most economical way to store large data sets.
Demands for archival and long-term storage are being driven by the ever increasing volume of data that home users and enterprises are creating and storing. This is leading to growth in external backup drives, NAS systems (local home storage servers), enterprise storage and cloud storage in general. This trend will increase the availability and accessibility of cloud and networked storage and may limit demand for storage capacity in personal devices.
For personal devices, the price gap between SSDs and HDDs is becoming smaller - reducing as new NAND technologies make it a more competitive storage option. In the enterprise market, the price per gigabyte for enterprise HDDs is falling at a similar rate as the cost per gigabyte for enterprise SSDs.
Spinning disks in the enterprise
Enterprise HDD capacity is expected to continue to grow with a potential to reach 20-40TB by the end of the decade, while still maintaining the industry standard 3.5-inch form factor.
Enterprise capacity HDDs with 8TB are already starting to answer the industry's nearline storage needs. New technologies such as Shingled Magnetic Recording (SMR) will increase the capacity on the same physical hardware - taking an 8TB drive to 10TB and further.
While conventional HDDs record data by writing to magnetic tracks that do not overlap, SMR HDDs record data on tracks that partially overlap like roof shingles and this provides a higher track density.
The increase in capacity is not without cost though the random write performance of these drives is limited and un-deterministic to a certain extent, while sequential R/W and random read performance is similar to the nearline models. This makes SMR drives suitable for the increasing need for streamed data, data archiving and cold data.
Smaller, faster, denser
While HDDs will still account for the majority of archival storage of the near future, their place within the server architecture is changing - more and more 15,000rpm drives are being replaced by enterprise SSDs.
There are a number of factors influencing this shift, the improved IO/s performance of eSSDs in comparison to even the fastest HDDs, the reduced energy consumption of flash-based drives, and the decreasing cost per gigabyte.
The cost per gigabyte is likely to continue to decrease, especially as the introduction of 3D NAND technologies such as Toshiba's 3D BiCS NAND helps to increase chip bit densities and increase drive capacities. While 3D technologies are still only just starting to make their mark on the storage landscape, SSDs manufactured using these new chips promise higher endurance, faster speeds, greater energy efficiency and higher capacities.
In the long term, SSDs have the potential to outgrow HDDs in terms of capacity, with 256TB SSDs predicted to become available within the next five years.
The changing face of interfaces
At the entry-level end of the enterprise market, SATA interface equipped eSSDs have proven very popular as boot drives and for larger storage volumes. One of the reasons for this is the fact that SATA SSDs provide the lowest cost/capacity entry point into the flash-based storage world. This economic advantage was driven by the high volume scale of client SATA SSDs being produced to replace 2.5-inch laptop HDD.
With additions such as power loss protection and increased over-provisioning, a client SATA SSD can be turned into a reasonably reliable and entry level eSSD. With a predicted decline of the 2.5" form-factor client SATA SSDs due to the transition to M.2 PCIe modules, this commercial driver will eventually disappear.
The use of new interfaces has accelerated the shift in form factor away from the 2.5-inch factor to M2 modules and even smaller. The world's smallest NVM Express SSD that squeezes 256GB into a single BGA package measuring just 16mm by 20mm was recently announced.
One of the limiting factors for the speeds at which SSDs can operate has been the way they interface with computers - the commonly used 6Gb SATA interface has a bandwidth of up to 600MB/s, while 12Gb SAS interfaces achieve up to 1,200MB/s. PCIe Generation 3 based SSDs typically support 4 lanes with bandwidths of approximately 1GB/s per lane, resulting in 4GB/s bandwidth. It is expected in the long term that the SATA eSSD market will transition to PCIe-based solutions (for boot) and SAS eSSD (for volume storage).
The higher bandwidth of PCIe enables the CPU to address more NAND chips at the same time, reducing latencies and enhancing R/W speeds. This has started a transition towards SSDs that have a direct connection to the PCIe bus using the NVMe interface protocol that has gained widespread adoption in many OSs.
The first server models that support this interface are available, but the number of devices is limited to the number of free PCIe connections in the motherboard's chipset. As PCIe-based enterprise SSDs of more than 4TB~8TB become available, this configuration will prove enough for tier 0 storage architectures. High DWPD PCIe SSDs will also become widely available in the Add-In-Card (AIC) form factor for direct plug in to the PCIe slots of the motherboard. While this approach reduces the need for cabling it, it does currently not allow any hot plug functionality - so removing or exchanging cards without service interruptions would not be possible.
It is important to note that hot plug capabilities were designed to overcome the unpredictable failure of individual HDDs in an array. As SSDs do not have any moving parts, their failure rates are far more predictable. Individual memory cells may wear out and fail, but in-built redundancy accommodates for this and monitoring software can identify drives that are reaching their endurance limits. From that perspective, failure rates are therefore similar than other semiconductor based devices in the server such as CPU or DRAM.
For larger, more complex storage systems, the 12Gb SAS interface is expected to remain the standard interface. The interface supports end-to-end data protection being provided by the SAS link, dual link capability and a large existing infrastructure of controllers, HBA, expanders, backplanes and JBOD.
To address differentiated applications and their different style of workloads, eSSD manufacturers are developing drives optimised for specific capacity points (such as 0.5TB, 1TB, 2TB, 4TB raw flash) and endurance levels (25, 10, 3 or 1 DWPD) to ensure the optimal balance of write endurance, capacity and cost for all applications and architectures.
RAID redundancy facing the chop?
There is a trend away from the use of traditional RAID array-based redundancy protection as continuously increasing drive capacities have made rebuild times unacceptably long for many organisations.
Instead, modern storage architectures based on alternative redundancy technologies such as Erasure Coding provide redundancy and data protection on a system level where several copies of data are stored on a number of individual servers. Depending on the system architecture, these systems can be resistant to individual drive failures, server or cluster failures, and even data centre failures.
Keeping an eye on industry drivers
The demand for storage capacity continues to grow as does the perceived value of the data stored. This is driving demand for cloud storage operations where economies of scale can help mitigate the costs of storing ever-increasing amounts of data.
While speed and capacity will remain important factors, power consumption and reliability will become even more important factors when determining TCO. Capacity per unit space is increasing in importance and high capacity, low endurance SSDs may become the industry's preferred solution.
In addition to cloud archival storage, the storage of video surveillance data is an important growth area. Applications for video surveillance systems keep expanding into many areas including industrial process monitoring, airports and traffic monitoring and a host of retail situations where customer behavior analysis can help improve overall customer experience.
Surveillance data is increasingly being used not only to protect against security risks, but also as a part of a comprehensive business process management tool.
Greater opportunities for enhanced video analytics are being offered with emerging, advanced video surveillance technologies and systems. The shift towards higher resolution cameras that have higher capacity demands together with the volume of video-data generated and the need to store data for longer, change storage requirements. This drives increased demands for local storage capacity as well as cloud-based video surveillance-as-a-service (VSaaS) offerings.
Here, the right, well thought through approach to storage is needed to make full use of the modern video surveillance solutions available on the market.
The adoption of new technologies and interfaces will continue to disrupt the existing perceptions of how and where HDDs and SSDs should be used in various storage architectures. The accessibility and growth of cloud storage and service offerings will dictate the need for storage in local devices for applications as diverse as file access, media streaming and surveillance storage and analytics.
No matter the application, the industry is reaching an inflection point in the use of HDDs switching to SSDs, and as the price per gigabyte falls, sales of SSDs will continue to grow and adoption will intensify and widen into new applications. What is clear, is that to be successful, manufacturers will need full control of design and development of drives and firmware - in the SSD market, that will by necessity include firmware, controllers and NAND flash memory cells.