VIDEO SDC EMEA 2022: Computational Storage: How do NVMe CS and SNIA CS work together?

[whoosh]

 

[ChatGPT]

SDC EMEA 2022: Computational Storage: How do NVMe CS and SNIA CS work together? In a detailed presentation at the SDC EMEA 2022, Bill Martin from Samsung delves into the intricate workings of Computational Storage (CS), focusing on how standards from NVMe and SNIA align and operate together. This topic is especially relevant as the demand for efficient data processing solutions continues to expand within storage technologies.

Overview of Computational Storage​

The concept of Computational Storage revolves around integrating data processing capabilities into storage systems. This presentation categorizes CS devices into three types:

• Computational Storage Processor: A standalone processor that accesses external storage but does not integrate storage itself.
• Computational Storage Drive: This device not only performs computational tasks but also contains onboard storage.
• Computational Storage Array: This setup incorporates multiple storage devices, either computational drives or traditional storage, managed by an array controller to function as a single computational entity.

Martin places particular emphasis on the Computational Storage Drive model, outlining its architectural elements. Key components discussed include:

• Computational Storage Engine: This is the core that performs computations, which can take various forms including dedicated hardware environments or virtual instances.
• Resource Repository: A central location for downloading or preloading environments and computational storage functions.
• Function Data Memory: Memory allocated for the execution of programs by the computational storage engine.
• Device Storage: Where data is physically stored, typically utilizing NAND flash technology.

NVMe Computational Storage Model​

The presentation also covers the NVMe's approach to computational storage, which encompasses components like:

• Compute Namespaces: Containing compute engines and associated programs, these namespaces are integral to managing and executing storage tasks.
• Subsystem Local Memory: This part includes program input, output data, and is where execution of tasks takes place.

Martin explains how terms between the SNIA and NVMe models translate, noting that while SNIA provides a broad framework adaptable to various protocols, NVMe's specifications cater specifically to its unique I/O commands.

Summary of Findings​

The takeaway from Martin's presentation highlights the importance of collaboration between SNIA and NVMe to streamline standards in computational storage. As the industry evolves, understanding these relationships and participating in standardization efforts through both organizations can foster advancements in technology. Through this presentation, professionals interested in computational storage are encouraged to explore not just the technical specifications but engage with related educational resources provided by SNIA and NVMe. The ongoing developments in this field hold potential for significant impact on data processing and storage efficiency. If you're curious about how computational storage might influence your work or have insights to share, let's discuss! What applications do you foresee benefiting from this technology?