The NGD Systems Newport: The Most Advanced Platform Today for NVMe Computational Storage

Hopefully our last blogs have helped you understand why it is difficult for applications to adopt new technologies. At NGD Systems, we have taken a “targeted adaptation” approach to enabling the use of computational storage by mainstream applications. These include the following:

• Use of a multi-core ARM processor as the heart of our computational storage platform. ARM processors are widely utilized and supported in our industry, with extremely healthy developer and 3^rd party development tool ecosystems.
• Use of standard Linux distribution (Ubuntu) and standard kernels. This allows our platform to run programs developed for X86 Linux platforms without modifying system calls.
• Use of a standard machine virtualization environment (Docker containers). This allows our platform to run programs in their own container, eliminating “interference” from other processes running on the system.
• NGD Systems’ management tools, which streamline the movement of data between processes running on the computational storage platform and those running on X86 processors.

These tools allow applications to be ported to the NGD Systems computational storage in hours or weeks, not months or years. As an example, NGD Systems, along with Microsoft, ported the Facebook Artificial Intelligence Similarity Search (FAISS) application to our Catalina-2 Computational Storage SSD in a little over a week. The impact on performance was dramatic – the time for FAISS to execute on the NGD platform barely increased as the size of the data sets increased, while FAISS running on a standard X86 platform saw dramatic increases in execution time as the size of the data set increased. FAISS running on the NGD Systems Catalina-2 barely exceeded an execution time of 0.5 seconds, while the time on the X86 for the largest data set was over 40 seconds. If you want to explore how NGD Systems can help to accelerate your applications, send me an email at eli.tiomkin@ngdsystems.com.