Tech has no shortage of buzzy new technologies -- and cutting through the hype to see what will actually impact the enterprise can be challenging. We're here to help. Starting in 2021, our contributors will give a rundown on an emerging tech and whether it'll pay off to pay attention to it. Here, we look at computational storage devices.
To see the other trends highlighted in our IT Trends To Watch series, read our Emerging IT Trends To Watch report.
What Is Computational Storage?
Computational storage is an architecture that moves compute operations onto the storage device itself rather than on external storage. The architecture allows data to be processed and analyzed where it is created and stored. The data remains in place.
Because computational storage eliminates the need to move data to external devices for processing, it can increase the performance of data-intensive storage applications.
How Does Computational Storage Work?
The two most common approaches to computational storage are general-purpose compute and stack optimization, according to Enrico Signoretti, senior data storage analyst at Gigaom.
In the general-purpose compute approach, a multi-core CPU or field-programmable gate array (FGPA) with RAM is integrated into the storage device. That integration allows data to be accessed concurrently by the host and integrated compute resources.
With stack optimization, compute resources in the storage device use APIs for software integration. Services like compression, encryption and data protection are integrated with the rest of the software stack to send high-demand operations to the device.
Computational storage doesn’t replace anything in a computing environment. Instead, the architecture moves compute operations onto the data storage device, bringing the compute operations close to the data. Computational storage allows for greater optimization.
How Long Has Computational Storage Been Around?
Computational storage emerged in 2011 when company-sponsored university research began to develop the architectural concept, said Scott Shadley, co-chair of the Storage Networking Industry Association (SNIA) Computational Storage Technical Working Group.
The architecture has had many names since its beginnings. SNIA and a group of founding member companies standardized the term computational storage in 2018.
Why Are People Paying Attention To Computational Storage?
Interest in computational storage has grown as organizations face more unstructured data and requirements to store and process data at the edge. “There is simply no way for the CPU and DRAM complexes of our current computing architecture to effectively manage all that data if it has been stored on a storage device,” Shadley said. “By allowing the [computational storage device] to manage localized data and only return useful data, the overall system can behave more effectively.”
The demand for computational storage devices will only intensify as the market understands that CPUs aren’t always the best way to process data, Shadley added.
Who Benefits From Computational Storage?
Computational storage has numerous uses, including the following:
- Solid-state devices (SSD): SSDs often suffer from slow writes. Computational storage can significantly improve performance.
- Edge computing environments: As data grows more quickly in edge environments, organizations run into limitations on space and power and need faster response times. Computational storage devices can act as the first line of compute on stored data at the edge. This can free up resources for localized compute and use the cloud or other system resources only for “after the fact” data management, Shadley explained.
- Content data networks (CDN): By moving computation closer to end users, computational storage can improve encryption, management and access to controlled content.
- Internet of things (IoT): Computational storage devices can optimize data ingestion, aggregation and tagging for IoT devices.
- Rich media processing: Typical methods of image processing require ingesting into the system and additional activity to read the image and extrapolate the data. Computational storage is a much more efficient method: When the image lands on the device, you can simply run the image software on that device, thereby freeing up the CPU, Signoretti said.
- Database acceleration: With powerful compute on storage devices, database operations and searches can be performed directly on the data, with no intermediaries.
- More efficient AI, machine learning and data analytics: Computational storage reduces data movement, scales easily, and improves parallelism in compression and decompression functions. As a result, it can free up GPU cycles to perform model training, analytics and other high-level functions.
Where Can You Get Computational Storage?
A growing number of vendors incorporate computational storage into their offerings.
- Computational storage drive: These devices, which typically come in the form of a solid-state drive, contain both compute and persistent storage. Vendors include NGD Systems, ScaleFlux, Samsung and Netint Technologies.
- Computational storage processor (CSP): CSPs are devices that contain a compute engine but do not have storage. Vendors include Eideticom and Pliops.
- Computational storage array: These products contain one or more compute engine and storage device but are managed as a system rather than a device. Nyriad is an example of a computational storage array vendor.