Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn how to activate untapped local NVMe storage within server clusters as a high-performance Tier 0 storage solution using Hammerspace technology in this 29-minute conference presentation. Discover how Hammerspace unifies local NVMe across server clusters into a protected, ultra-fast storage tier that accelerates AI workloads and increases GPU utilization while reducing infrastructure costs. Explore the core problem of stranded local NVMe capacity in servers that remains underutilized despite its speed, and understand how network bottlenecks limit AI workflows with growing context lengths and fast token access requirements. Examine Hammerspace's standards-based approach using NFS, SMB, and S3 protocols without requiring client-side software installations, leveraging parallel NFS v4.2 with flex files contributed to the Linux kernel. Understand how the technology transforms compute nodes into storage servers without creating isolated islands, enabling programmatic data orchestration across tiers and cloud storage. Review performance benefits through theoretical results and MLPerf benchmarks demonstrating superior performance per rack unit, along with Hammerspace's contributions to Linux kernel enhancements for reduced CPU utilization and accelerated write performance. Gain insights into the non-disruptive architecture that allows live data mobility while maintaining seamless user experience, effectively unlocking petabytes of previously unused storage capacity.
