NVM Express has introduced the NVM Express (NVMe) Computational Storage Feature. This innovative development establishes a standardized, vendor-neutral framework that bridges applications with NVMe Computational Storage devices, integrating both compute and storage services. This launch represents a significant advancement in the way enterprises and hyperscale data centers manage and process data.
Bill Martin, Co-Chair of the NVMe Computational Storage Task Group and a Board Member, highlighted the strategic importance of this release. “Our introduction of NVM Express Computational Storage marks a pivotal step in adapting to the dynamic demands of the storage industry. This standardized approach fosters an open, interoperable ecosystem, offering the potential for reduced ownership costs and enhanced performance through computational offloading.”
The essence of Computational Storage would lie in its ability to minimize data movement by enabling computation directly on the storage device. This innovation would particularly be beneficial for latency-sensitive applications, such as databases, AI suggestion models, and content delivery networks, where faster response times are crucial.
Modular Approach
To sum up, the NVM Express Computational Storage Feature introduces two groundbreaking command sets:
- The Computational Programs Command Set offers a host-driven, modular framework for computational programs, enabling more efficient data processing.
- The Subsystem Local Memory Command Set allows for accessing memory within an NVM subsystem. This is facilitated through computational programs and NVMe transport, which are accessible via NVMe I/O commands.
When combined, these new features would represent a significant advancement in computational storage technology, with the potential to improve data processing performance and efficiency in a variety of sectors. This innovation is anticipated to have a significant impact on how data processing and storage are developed in the future, providing real performance, cost-effectiveness, and scalability advantages to enterprises and data centers.
