Introducing Tier 0 w/ David Flynn

This Data Unchained podcast episode interviews Hammerspace CEO David Flynn about their new Tier 0 capability for global data platforms. The discussion centers on solving the problem of underutilized NVMe storage in GPU servers used for AI and HPC, explaining how Hammerspace's technology aggregates this local storage into a shared, high-performance pool without requiring additional software or hardware. Flynn details a customer success story demonstrating rapid deployment and significant performance gains. A key takeaway is Hammerspace's "local IO patch set," which allows near-local NVMe access speeds even within a shared filesystem, minimizing latency and maximizing GPU utilization. The interview concludes with a call to action to visit Hammerspace's booth at the upcoming Supercomputing conference.

Outlines

Part 1: Introduction and Problem Definition

Part 2: Hammerspace Solution and Optimization

Part 3: Differentiation and Reliability

Part 4: Deployment and Conclusion

Sign in to continue reading, translating and more.

Open full episode in Podwise

Hammerspace

Part 1: Introduction and Problem Definition

Introduction to Data Unchained and Guest Introduction

David Flynn's Background and the Problem of Trapped Capacity in GPU Servers

Part 2: Hammerspace Solution and Optimization

The Hammerspace Solution: Tier 0 and Hyperscale NAS

Customer Success Story and Time-to-Value

Hammerspace's Approach to Performance Optimization: Minimizing Latency

Data Protection and Tiered Storage

Addressing the "Idle GPU is a Sin" Problem with Checkpointing

Part 3: Differentiation and Reliability

Differentiating Hammerspace from Hyper-Converged Infrastructure (HCI)

Addressing Reliability Concerns in a Distributed Environment

Overcoming the Noisy Neighbor Problem with Kernel-Based NFS

Part 4: Deployment and Conclusion

Deployment Scenario and Call to Action

Introducing Tier 0 w/ David Flynn

Hammerspace

Part 1: Introduction and Problem Definition

00:09Introduction to Data Unchained and Guest Introduction

Introduction to Data Unchained and Guest Introduction

01:26David Flynn's Background and the Problem of Trapped Capacity in GPU Servers

David Flynn's Background and the Problem of Trapped Capacity in GPU Servers

Part 2: Hammerspace Solution and Optimization

04:11The Hammerspace Solution: Tier 0 and Hyperscale NAS

The Hammerspace Solution: Tier 0 and Hyperscale NAS

07:03Customer Success Story and Time-to-Value

Customer Success Story and Time-to-Value

11:06Hammerspace's Approach to Performance Optimization: Minimizing Latency

Hammerspace's Approach to Performance Optimization: Minimizing Latency

13:18Data Protection and Tiered Storage

Data Protection and Tiered Storage

17:19Addressing the "Idle GPU is a Sin" Problem with Checkpointing

Addressing the "Idle GPU is a Sin" Problem with Checkpointing

Part 3: Differentiation and Reliability

21:17Differentiating Hammerspace from Hyper-Converged Infrastructure (HCI)

Differentiating Hammerspace from Hyper-Converged Infrastructure (HCI)

24:30Addressing Reliability Concerns in a Distributed Environment

Addressing Reliability Concerns in a Distributed Environment

28:34Overcoming the Noisy Neighbor Problem with Kernel-Based NFS

Overcoming the Noisy Neighbor Problem with Kernel-Based NFS

Part 4: Deployment and Conclusion

33:04Deployment Scenario and Call to Action

Deployment Scenario and Call to Action