By: Ravi Thummarukudy, CEO Mobiveil, Inc.
Facebook contends that “two billion photos are shared daily on Facebook services,” a huge amount of data that must be stored, indexed, and managed. However, this will be dwarfed by the data tsunami created by the Internet of Things (IoT). In the article “Making sense of Seagate’s decision to tackle the Data Centre Interconnect market,” author Desire Athow writes that “by 2020, 44ZB of data will be created, of which 13ZB will need to be stored, however the amount of data that installed capacity will be able to hold will only be 6.5ZB.”
To meet the challenge of accommodating the data tidal wave hurtling toward the data center, data center architects began embracing virtualization and distributed computing using commodity components, running Linux and an open source variant of Unix to deliver enterprise-class stability and performance. This trend is best exemplified by huge retailer Amazon and social media giant Facebook among others.
While the virtualization of the server and storage farms is occurring, another transformation is occurring in the semiconductor devices that make up the server and storage farm hardware. This change is required by the need for bandwidth and intelligence to move the data in and out of storage. The first of the bottlenecks is the hard drives, millions of rotating disks drive in large storage farms, that hold the vast majority of all data collected today. Each drive stores bits of data in concentric tracks on a magnetic disk surface with a mechanical actuator positioning a read/write head over the track to store and extract data from the track. Access times to the data stored on these surfaces are on the order of milliseconds.
In his “Fast Company” article, “How One Second Could Cost Amazon $1.6 Billion In Sales,” author Kit Eaton stated, “Amazon calculated that a page load slowdown of just one second could cost it $1.6 billion in sales each year. Google has calculated that by slowing its search results by just four tenths of a second they could lose 8 million searches per day—meaning they’d serve up many millions fewer online adverts.”
To break the bottleneck in these data centers demands improving the performance between the servers and large storage farms. The ever-decreasing cost of flash memory made it cost effective in recent times to build high capacity solid-state drives (SSDs) to act as a cache hiding the latency of mechanical storage devices. By moving frequently accessed data out of rotating storage and semiconductor flash memory chips, data can be accessed in microseconds, thus tremendously reducing transaction times and improving the customer experience.
A Carnegie Mellon study found that for I/O-intensive workloads such as cp, PostMark, and MySQL, NVM provided an order of magnitude improvement in performance. For data centers doing e-commerce, Internet searches, fraud detection, and other data mining operations, this performance boost has a huge positive impact on their operation.
However, one of the problems solid state drives encountered was the limitation of the interface between the drive and its controller electronics. SSDs employed the same interfaces used for rotating memory devices. Since the latter had built in mechanical limitations, the bandwidth expectations of this interface was not as great as that for solid state storage devices that could move large amounts of data much faster. To break this bottleneck, Intel Corp. led a group that formed the NVM Express Organization and the group released the NVMe specification on October 11, 2012.
Mobiveil was quick to develop NVMe intellectual property (IP) that met the requirements of the standard. In addition to NVMe, the company had also developed a state-of-the-art enterprise flash controller and a Peripheral Component Interconnect Express (PCIe) IP block along with Low-density parity-check (LDPC) error correcting software for flash memory. By providing this IP portfolio, designers building chips to use the NVMe interface could license the all these elements needed in his NVMe flash controller design rather than having to create and maintain them. To date, the company’s IP portfolio has been designed into multiple chips installed in data centers.
In addition, NVMe is also creating a unique opportunity for data center engineers to customize their system solutions to more precisely address the bottlenecks in their data center environment. For example, a data center primarily designed as an Internet search engine will see a different pattern of disk accesses than one primarily aimed at on-line shopping, or one serving multiplayer games.
In light of this dynamic, it is not difficult to understand Intel’s $16.7B purchase of Altera. Data Center engineers who previously purchased packaged chip solutions are now beginning to program in languages used to design semiconductor chips. However, rather than fabricate silicon from the code, the program is loaded into Field Programmable Gate Arrays (FPGAs) supplied by Altera and Xilinx. Not only are the FPGAs highly optimized to accelerate each data centers’ operation, the solution also provides competitive advantage.
For data center engineers wanting to build their own FPGA solution, Mobiveil has put together a complete platform—a printed circuit board with Altera FPGA. The board contains all the standard hardware and software elements that enable data center engineers an accelerated path to product development. With the platform, the engineer only needs to add the custom program he creates for his application into the FPGA and plug the board into his server.
Data Centers of today are demanding the ability to rapidly and flexibly evolve. Furthermore, in a 24/7 always-on world, downtime is not an option. Thus, the data center needs to grow and evolve, while continuing to serve its online customers. The application of programmable chips into the flow of data between server and data farm has the potential to meet this rigorous requirement. As the architecture of the data center evolves over time, data center engineers simple reprogram their proprietary algorithms, remove the old code in the FPGA and replace it with the new code. Likewise, as the NVMe interface standard evolves over time to accommodate faster data rates, Mobiveil can easily upgrade the platforms containing its IP with simple software update.