SDXI is a memory to memory data movement and acceleration standard developed by SNIA's SDXI Technical Working Group(TWG). Join this community outreach open to all to see if your use cases are covered by the SDXI TWG. Bring your ideas, pain points, insights to collaborate.

Customers are raising concerns about permission handling and security for multiprotocol access of data shared across NFS, SMB and S3 or any combination thereof especially with AI models analyzing data.

The Distributed-NVMe is my personal open source project. It is a distributed block storage system. It provides the standard NVMe-oF interface to the hosts. A single virtual block disk could provide more than 2 millions IOPS in the TCP NVMe-oF mode without any NVMe-oF specific hardware offload. Homepage: https://github.com/distributed-nvme/distributed-nvme

This keynote will provide an in-depth overview of Ultra Ethernet.

The massive growth of hyperscale AI has pushed the HPC-AI market to a critical inflection point in 2025. The market is increasingly defined by companies spending tens of billions of dollars per year, building massive data centers hundreds of megawatts — even gigawatts — at a time. As a result, national sovereignty sensitivity is on the rise, and we see independent technology tracks from the U.S., China, Japan, the EU, the UK, India, and others. This has significant implications throughout the industry, including for storage, networking, and data management. 

The AI Pipeline is a complex set of phases and operations, all with different requirements for the underlying storage which affect storage technology choices. Understanding this complexity can be overwhelming and hard to understand. This presentation will provide a 101 style view of the typical demands of the AI phases on the storage which impacts your storage choices and success deploying your AI strategy. Storage is critical for efficient AI, and is currently overlooked, to some extent, by the market.

In this talk, the author will explore the use of accelerators for the training and inference of models, and look at the proliferation of methods and models in the AI world.

As with any emerging technology, developing open-source software for CXL is challenging and emulation of features is a key path forward. In this talk, we will introduce the current major CXL features that QEMU can emulate and walk you through how to set up a Linux + QEMU CXL environment to enable testing and developing new CXL features. Some limitations exist, as with any platform, which we will discuss along with the latest developments in support for dynamic capacity devices (DCD) and switches.

The migration of a VM and the attached local NVM Express™ SSDs from one server to a different server does not provide a standard mechanism to guarantee that the information returned to the VM by Admin commands after the migration will be changed due to the different SSDs.