Intelligent edge computing applications have gained major momentum recently, while hyperscale system growth has accelerated. Each offers tremendous possibilities that can transform technology. However, both rely on fundamentally old computing architectures. VMblog recently spoke with industry expert, Scott Shadley, Principal Technologist at NGD Systems, to learn more about this area of the industry.
VMblog: There's
been some buzz about computational storage from a few different perspectives.
How does NGD define computational storage and why is your approach different?
Scott Shadley: NGD Systems' approach
to computational storage centers on In-Situ processing. In-Situ processing is
processing that's done right where the data resides. NGD Systems brings
computational resources right to the storage device (in our case, an SSD).
NGD Systems unites
compute and storage in the highest capacity and most power efficient NVMe SSD
available in the industry's smallest form factor. This patented technology can
save enterprises up to 45 percent annually in costs associated with physical
footprint, server expense and energy. Newport Platform of NVMe computational
storage solution provides increased efficiency by running applications like AI,
via In-Situ processing, for mass datasets wherever the data is generated and
stored. This radically reduces the network bandwidth required, by up to
500 percent, to analyze mass data sets produced for AI and other data analytic
applications.
NGD Systems' CEO,
Nader Salessi, was the pioneer behind In-Situ processing and NGD Systems is
credited as the first company to deliver In-Situ processing technology in a
commercially available platform.
VMblog: Why
is data movement so cumbersome?
Shadley: Data naturally has
gravity and requires resources (host memory and CPUs) and energy to move. As
deployments grow, data is moved increasingly longer distances between nodes and
local compute/memory complexes, increasing resource and energy usage, and thus
costs.
Until recently, the
size of typical data sets has made data movement only moderately costly.
However, as data sets grow and data-intensive applications such as Big Data
analytics, artificial intelligence (AI), machine learning (ML), genomics, and
IoT gain in use, the costs and time needed for data movement is becoming
critically challenged. Moving massive amounts of data from storage to host CPU
memory to process a query is costly in terms of power consumption and time.
The impact of data
movement is being felt in nearly all compute applications. Even in consumer
devices such as smartphones, tablets, mobile-PCs, and wearable devices, where
cloud services are becoming necessary, it has been shown that data movement
between the main memory system and computation units accounts for, on average,
62.7% of the total system energy. For Big Data, AI and machine learning
applications with large data stores and significant search, indexing, or
pattern matching Workloads, the cost of
data movement is even greater.
VMblog: What
other ways have vendors tried to address this problem?
Shadley: Vendors have attempted
to address the challenge of data movement by delivering disaggregated
solutions, such as NVMe-oF fabrics, composable architectures and GPU and FPGA
accelerators. While these can speed up the process to some degree, they don't
move eliminate the data movement challenge and only minimize some of its
effects. All these solutions require space and power needs that may not exist,
and they are not innovating the way to move and manage the stored data itself.
VMblog: What
sort of companies are looking to deploy computational storage technology?
Shadley: Computational storage
is ideal for any organization employing hyperscale environments, edge
computing, content delivery networks (CDNs). For example, this would include
web companies like Facebook, cloud providers like AWS, telcos like AT&T and
CDNs like Akamai. Ultimately, though, computational storage is useful for any
company relying heavily on AI and data analytic applications.
VMblog: What
are some of the real-world applications that computational storage supports -
now or in the future?
Shadley: Computational storage
enables a wide variety of compelling use cases. This is especially apparent in
edge computing and IoT. Imagine a commercial jet that uses sensor technology to
determine in seconds rather than hours what its maintenance needs are as it
sits outside the gate before its next takeoff. Computational storage would be
needed to provide such efficiency and power in a small form factor.
Another great example
of an edge implementation is object tracking in surveillance. Consider a remote
camera platform that can analyze and track a single person in a stadium in real
time by running the AI-based search algorithm while the data is being stored on
cameras. No need to ‘look back' over the data.
VMblog: And finally, are
there any key partners or industry groups helping advance computational
storage?
Shadley: The Storage Networking
Industry Association (SNIA) recently launched the Computational
Storage Technical Working Group. This SNIA body is focused on developing
standards to promote the interoperability of computational storage devices, and
on defining interface standards for system deployment, provisioning,
management, and security. These efforts will enable storage architectures
and software to be integrated with computation in its many forms.
##