Over the past year, we’ve seen that HDD capacity increases have plateaued. Spinning disks have been surpassed in storage capacity by SSDs. Performance comparisons between the two is not even a topic of debate. For CIOs, the deployment of flash storage arrays is easy and offers an immediate boost in IOPs for critical applications. More importantly, all the innovation in new drive development has shifted to flash. We are now seeing many approaches being tried in the market to boost SSD performance even further, to scale up to new drive capacities and new array architectures, to adopt new form factors for better rack-scale integration, and increase manufacturing volume to finally meet market demand.
In the first part of this article, we covered Samsung’s rapid progression with 3-D NAND technology. With the arrival of its 5th generation 3D NAND next year we will see 2.5” SSDs soar into the 128TB range. The company says its on-track for 5 more generations of 3-D NAND in the coming decade. In this second part of the article, we’ll look at innovations from another giant, Intel, which has also set its sights on bringing non-volatile memory technologies to the forefront of server, system and data centre design, as well as developments from Nimbus Data and the Gen-Z consortium.
Intel’s non-volatile memory advancements
Intel began shipping its first SSDs as early as 2008 and has been on a continuous improvement path ever since. In 2010, Intel and Micron Technology entered into a partnership focused on NAND flash memory.
In 2015, Intel and Micron announced 3D XPoint technology, which was described as the first new memory category since the introduction of NAND flash in 1989, with promises to be up to 1,000 times faster and up to 1,000 times greater endurance than NAND, while being cheaper than DRAM and non-volatile. Intel then adopted the "Optane" brand for products based on thistechnology, while Micron adopted the QuantX brand. Optane is fundamentally different from NAND and uses a combination of unique Intel memory + storage controllers, Intel interconnect IP, and Intel software.
Introducing the memory ruler
Intel's big news at last week's Flash Memory Summit was its new "ruler" form factor for SSD. Instead of the
traditional, 2.5" or 3.5" rectangular box for disk drives, Intel's ruler is a long, thin box designed to slide in to a 1" server chassis, plugging in via a PCIe interface at the end of the ruler. It is a slick design. Apart from
looking better, the long, thin shape dissipates heat easier. Intel showed a 1” RU server chassis accommodating
32 of these SSD rulers, creating up to 1 petabyte of storage. Intel could offer Optane SSDs and/or 3D NAND
SSDs in this form factor.
It’s been a while since a new storage drive format gained widespread acceptance. Intel will need to bring its new form factor to standardization, perhaps via the Open Compute Project, although this was not confirmed. The ruler design should prove to be particularly useful in hyperscale data centres, where plug-n-play convenience is especially useful when 100s of thousands of servers need to be maintained. Intel also noted that its ruler form factor could be used for plug-in accelerators, perhaps FPGA boards optimized for specific functions. No timeline was given for when the ruler might enter the market.
Intel and Attala Systems also announced an FPGA-based accelerated RDMA over Converged Ethernet (RoCE) networking solution designed to serve as high-performance, composable storage infrastructure with features such as self-learning orchestration and provisioning capabilities. The idea is to create an adaptable storage infrastructure that is essentially an elastic block storage (EBS) solution, accelerated. Attala Systems is a start-up based in San Jose, California that was founded by Sujith Arramreddy, who previously co-founded ServerEngines (acquired by Emulex for $250 million in 2010) and ServerWorks (acquired by Broadcom for $1.4 billion in 2001). Attala's CEO is Taufik Ma, who previously was co-GM of Intel's Server System business unit before leaving for a storage/networking start-up. Nimbus Data sees 500 TB SSDs by 2020
Nimbus Data is a privately-held develop of all-flash arrays based in Irvine, California. The company observes that 40 million nearline/high-capacity HDDs are shipped per year, and all of them use the 3.5” form factor. At Flash Memory Summit, Nimbus Data introduced a software and multiprocessor solution for OEMs developing next-generation solid state drives for data centres. Whereas conventional SSDs are based on a single flash controller, Nimbus ExaDrive is based on a distributed multiprocessor architecture. Inside an ExaDrive-powered SSD, multiple ultra-low power ASICs exclusively handle error correction, while an intelligent flash processor provides wear-leveling and capacity management in software. Nimbus sees an opportunity for its ExaDrive being used in super capacity SSDs that let data centers rip-andreplace HDDs with flash. ExaDrive supports the standard SAS interface and is optimized to fully utilize the volume of the 3.5” form factor.
Nimbus said its ExaDrive is used by Viking Technology and SMART Modular Technologies in 50 TB and 25 TB SSDs for cloud infrastructure, technical computing, and digital content storage. The company predicts that its ExaDrive software-defined architecture will enable SSDs as large as 500 TB by the year 2020, achieving up to 600 petabytes in a single rack. This represents a 50x increase over what is possible with HDDs today. “ExaDrive’s software-defined multiprocessor architecture for SSDs delivers a game-changing leap forward in capacity, density, and energy efficiency that HDDs will never be able to recoup,” stated Thomas Isakovich, CEO and Founder of Nimbus Data. “ExaDrive broadens the appeal of flash memory to tier 2 and nearline use cases, enabling flash to become the dominant data center storage media.”
Gen-Z consortium targets data centres
The Gen-Z Consortium is a vendor-led group that is developing an open systems interconnect with memory semantic access to data and devices via direct-attached, switched or fabric topologies. Its major members include AMD, ARM, Broadcom, Cray, Dell EMC, Hewlett Packard Enterprise, Huawei, IDT, Micron, Samsung, SK hynix, and Xilinx. At this year’s Flash Memory Summit, the group had planned it’s the Gen-Z multi-vendor technology demonstration, connecting compute, memory, and I/O devices. Despite the unfortunate fire at a vendor booth on the opening day of the event, the demo was still able to occur in a nearby meeting room.
The demo showed FPGA-based Gen-Z adapters connecting compute nodes to memory pools through a Gen-Z switch, creating a fabric connecting multiple server vendors and a variety of memory vendors. Such a highperformance and scalable fabric/interconnect could be implemented in future data centres. The demo also featured a scalable prototype connector defined by the Gen-Z Consortium, running at 112 giga-transfers/sec. “We are excited to showcase the first technology demonstration of Gen-Z that includes solutions from multiple member companies, including a variety of servers, memory and I/O devices, all connected with a Gen-Z fabric,” said Kurtis Bowman, President of the Gen-Z Consortium. “The consortium continues to meet the planned development schedule and we expect to see initial Gen-Z products in the 2019-2020 timeframe.”
http://genzconsortium.org/
