An informal gathering has given birth to a new working group that will tackle challenges around computational storage.

The Computational Storage Technical Work Group (TWG) launched by the Storage Networking Industry Association (SNIA) will bring together member companies to create standards to promote the interoperability of computational storage devices and to define interface standards for system deployment, provisioning, management, and security.

Tom Friend, the group’s co-chair as well as director of industry standards at SK Hynix, said that the idea for the group came from an informal early-morning “birds-of-a-feather” session at the recent Flash Memory Summit. It ended up overflowing a large room of people from across the storage industry with multiple representatives from large companies. “Not all of them were in storage or compute,” he said. “It seemed to span across a very wide group of disciplines. That told me that we must be on to something.”

Founding members in the Computational Storage TWG include Arm, Lenovo, Micron Technology, Inc., NetApp, Samsung Electronics Co. Ltd., Western Digital Corporation, and Xilinx, among others, and any company that joins SNIA is eligible to participate in this group as well as its 14 other technical work groups.

Some of the companies in the Computational Storage TWG are startups, added co-chair Scott Shadley, which have been looking at storage from a different angle as chips get bigger — more than just storing bits.

“We all see a change in storage coming,” Shadley said. It made sense to propose the working group under the SNIA umbrella because it’s storage-focused and interface-agnostic.

“Things just don’t get adoption in the industry unless they’re standardized,” said Shadley, and the group will be reaching out to existing standards organizations such as PCIe and NVM Express to participate as they enable the shift to heterogenous compute architectures close to storage data workloads.

“The idea is not to create something new but to help make sure that whatever exists is augmented in the right way,” Shadley added. This includes everything from persistent memory to networking-centric storage architectures. “We’ve also got a very broad range of interests from the operating system and the software world as well.”

Shadley said that one overarching goal is to demonstrate that devices can do more than just store bits and how to manage those devices in a way that’s open, easy, and implementable or already part of another standard, such as the open NVMe protocol. “You’re able to eliminate some of the steps and hops and free up the applications and software to actually see what the products can do or are capable of.”

Another goal is to further move the compute closer to storage to alleviate the bottlenecks being created as drives handle larger data sets and overrun the memory footprint. Computational storage solutions generally focus on applications in which the demand to process ever-growing storage workloads is outpacing traditional compute server architectures, such as artificial intelligence, big data, content delivery, databases, and machine learning.

The immediate step for the group over the course of the next year is to figure out which problems it will focus on solving and those that it won’t. “There is a lot of moving parts and there is a lot of ways that this can move forward,” Shadley said.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved