Creating a High-Performance Standard for AI Workloads

The increasing importance of AI in the enterprise has transformed the management, movement, and storage of unstructured data. Enterprises now require the high levels of storage performance and scale previously reserved for elite HPC and Hyperscale environments, as well as advanced data services to find, stage, and process the data needed for GenAI and inference workloads.

The problem is that AI workloads require high-performance access to datasets that are often distributed across existing on-prem and cloud-based storage silos, requiring performance levels traditional enterprise IT architectures are not designed for.

But within every enterprise data center there are Linux-based servers that already contain a series of high-performance enhancements that help accelerate existing IT environments to meet these challenges.

With an ongoing series of updates to the Linux NFS kernel and pNFSv4.2, standard distributions now include key features for high-performance workloads and the distributed hybrid storage environments needed for AI and other GPU-based use cases. These enhancements enable modern data centers and clouds to support such workloads globally with standard Linux, and without the need for proprietary clients to support multi-vendor data environments.

In this session we will cover the latest kernel additions up through Kernel 6.18, to enable such use cases, and look ahead to future enhancements coming in the next year.