Faster flash is taking over data centers (Part 2)

Posted on 04/18/2018 by

Anders Graham

This is the second post in a two part series on Flash memory (read part 1 here). Used in billions of consumer cellphones to store photos, video and music, and almost ubiquitous in today’s lightning-fast and ultra-thin PC designs, Flash memory is now rapidly displacing traditional hard disk drives (HDDs) as the storage medium of choice for data centers.

We will soon see so-called “performance” HDDs relegated to the history books, replaced by Flash memory, while much slower, cheaper HDDs will be consigned to housing data that very rarely needs to be accessed by its owners. Although this “cold data” will continue to grow as humans (and machines) generate new digital content at an astounding rate, the more fun, interesting, and actionable data, sometimes called “hot”, “warm” or even “luke-warm” data, is moving resolutely to Flash.

Flash technology is already much faster and less power hungry than HDDs. It is also becoming increasingly compact, allowing scores of densely packed Flash chips to be squeezed into solid state drives (SSDs), which already can hold more than twice as much data as the highest capacity HDDs. We’re witnessing a stunning transformation in data storage as Flash memory takes over an increasing percentage of data center workloads. And Samsung – the world’s leading producer of Flash memory – is leading the way in several key technology areas, including:

 

1. Ridiculously dense solid state storage

Back in 2013, Samsung announced its first use of 3D vertical NAND (V-NAND) technology – which used chips containing up to 24 layers of vertically stacked memory cells – to produce enterprise SSDs that could store nearly a terabyte (TB) of data. We shipped the first 15.36 TB SSDs using this technology in early 2016, and then upped the ante in early 2018 to 30.72 TB SSDs. These latest SSDs not only offer more than twice the capacity of the largest available hard disk drives (HDDs), they do it in a more compact form factor (drive casing) that occupies roughly half as much space.

 Samsung recently invented an even more compact form factor for next generation enterprise SSDs, called the NF1, which you can learn more about by reading this product brief or by watching this quick video clip.  A 1.75-inches high “1U” server rack can hold up to 36 of these NF1 SSDs. That’s triple the capacity of previous server configurations, which typically required a 3.5-inches high “2U” server rack to hold just 24 SSDs.

 

2. High performance PCIe NVMe storage

Samsung also has been at the forefront of developing Flash drives that combine the increasingly popular, non-volatile memory express (NVMe) storage protocol with the high-speed PCI Express (PCIe) interface to accelerate data transfers between SSDs and central processing units (CPUs). NVMe was built from the ground up to maximize Flash storage performance, replacing legacy disk-based protocols (such as AHCI), and uses the PCIe interface to enable a high bandwidth connection to the computer. For example, consider that an NVMe SSD connected over 4 PCIe lanes (the most common configuration) at 8 gigabytes per second (Gb/s) per lane results in over 5 times the bandwidth of a legacy SATA SSD, which only uses 1 lane at 6 Gb/s (4×8=32 Gb/s compared to 1×6=6 Gb/s). So, the software is designed for Flash, and the hardware is designed for max data transfer capability. This results in a tremendous performance improvement when moving to this new technology.

Flash-optimized NVMe SSDs are expected to become the most popular SSDs for PCs later this year, as computer users continue their shift from HDDs to Flash drives, which offer higher-performance, always-on functionality that can fit into thinner, sleeker laptop designs. Samsung has been the leading supplier of enterprise SSDs across all interfaces, including NVMe, since late 2016.

Meanwhile, cloud storage providers have been jumping on the NVMe train, and are now rolling out NVMe-based server designs throughout their data centers. At this year’s Open Compute Platform Summit, virtually all of the new server designs shared by major cloud players featured NVMe SSDs.

Samsung’s PM963 SSD, launched in 2016 to facilitate this wave of cloud storage innovation, is probably (based on analyst reports and Samsung’s own data) the world’s best selling NVMe data center SSD. Comprised of state-of-the-art Samsung TLC Vertical NAND Flash memory, it is optimized for data center environments and engineered for outstanding performance under varying data center workloads. PM963 uses four PCIe lanes to deliver approximately four times the performance of SATA SSDs for sequential and random read, enabling significantly better performance-to-power efficiency compared to SATA. Our next-generation NVMe data center-class product, the PM983, which will offer up to 15.36TB of storage in the new NF1  form factor, should provide the best of both worlds: high storage density and significantly faster data transfers.

But what if you could also reduce the time required to fetch data from the SSD? Wouldn’t that be truly a game changer?

 

3. Low Latency SSDs

Samsung recently announced mass production of a new “low-latency” memory product called Z-SSD, which uses a proprietary vertical NAND technology to maximize the data bandwidth available through a PCIe interface. More importantly, its 16 microseconds of latency, the wait time required to execute a request, is five times less than a typical NVMe SSD. Samsung is working with partners to understand the benefits this can provide, but lower latency means faster access to stored data, which could be a valuable advantage for those involved with real-time data analysis and other high-performance computing tasks.

Early research suggests that Z-SSD and similar technologies could have a profound impact by reducing storage latencies in relational databases, and they could also be used in applications such as controller caching in all-Flash storage arrays. Eventually, I would expect Z-SSD to be implemented as an entirely new tier of high-performance, low-latency storage.

So what do dense storage, high performance, and low latency have in common? You guessed it – solid state storage. And as data center demand continues to soar, we can expect Samsung and its industry peers to continue accelerating the pace of innovation. A plethora of interesting uses are already being considered for next-generation SSDs. But that will have to wait for another blog… So for now, buckle up!

Tags: , , ,

Leave a Reply

Your email address will not be published. Required fields are marked *

Captcha *