Computational storage is an innovative technology that combines processor and storage functions within a single device. "It's a storage subsystem that includes a number of CPUs located on the storage media or their controllers," says Johan Alexander, CEO of app development firm APKCima.
"Think of it as a mini server built into the drive," says Cindy LaChapelle, a consultant with technology research and advisory firm ISG. "The ability of computational storage to move data handling, scanning, and data processing closer to the data source via direct integration of processing capability into the storage device significantly reduces network traffic and power usage, lowers latency and improves scalable performance," she explains.
The Storage Networking Industry Association (SNIA) defines computational storage as "architectures that provide computational storage functions (CSFs) coupled to storage, offloading host processing or reducing data movement." The advantage of computational storage over traditional storage, LaChapelle notes, is that it pushes the computational requirement to handle data queries and processing closer to the data, thereby reducing network traffic and offloading work from compute CPUs.
There are two general categories of computational storage: fixed computational storage services (FCSS); and programmable computational storage services (PCSS). "FCSS are optimized for specific, computationally intensive tasks such as inline compression of encryption at the drive," LaChapelle says. "PCSS can host an operating system, typically Linux-based, and are dynamically programmable by the end user."
There are several different approaches to computational storage, such as the integration of processing power into individual drives (in-situ processing), and accelerators that sit on the storage bus at the storage controller, not in the drives themselves. "Most computational storage solutions leverage NMV-Express SSDs as the base storage technology," LaChapelle says.
Initial Computational Storage Adopters
Most early computational storage adopters are organizations that deal with massive amounts of data, yet also require high-performance computing. Such organizations include cloud service providers, data center operators, and artificial intelligence developers, Alexander notes. "The companies or industries who have adopted this technology [include] Tesla, Google, Facebook, and Yahoo Mail."
Hyperscalers such as Alibaba and AWS have also embraced computational storage, LaChapelle says. She observes that mainstream storage vendors are also jumping on the bandwagon, adopting fundamental computational storage concepts, such as in-line compression and encryption performed at the storage device or controller level. "As the technology evolves, it will ultimately lead to applications being able to support computational storage out-of-the box," LaChapelle predicts, adding that "the technology is not there yet."