This is an SSD?! - PureStorage FlashBlade Tour

This video presents a deep dive into Pure Storage’s FlashBlade architecture, framed around a hands-on tour of their data center facilities. It explains why SSDs in large-scale deployments still face bottlenecks, and how Pure Storage rethinks flash storage by moving logic and control closer to the data itself. The presenter introduces the concept of direct flash modules, or DFMs, which are larger than typical SSDs and connect to a blade-based chassis that hosts multiple DFMs. The discussion emphasizes how the absence of traditional DRAM caching inside each module is compensated by software-driven data management, overprovisioning, and rapid parity-based data reconstruction, enabling fast, consistent performance at scale. The narrative highlights the critical role of software in orchestrating a fabric of interconnected blades, the modular power and cooling design, and the evergreen software model that enables feature upgrades without downtime. By contrasting QLC with TLC memory and explaining the unique control plane, the host clarifies how Pure Storage achieves high efficiency and reliability in enterprise workloads, including AI and deep learning pipelines. In the middle segment, the focus shifts to the FlashBlade//S architecture, where multiple blades connect via a high-speed fabric using FIOMs to form a single, scalable storage system. The video demonstrates the density and power efficiency advantages of flash over spinning disks, noting how a single 5U chassis can host substantial capacity and redundancy. The host also explains how the system uses a software-defined file system that behaves like a database with authorities, enabling object stores similar to Amazon S3, SMB, and NFS access. Practical demonstrations show how a DGX A100 can be fed with data from the FlashBlade array, underscoring the partnership between Pure Storage and NVIDIA for AI workloads. The discussion makes a strong case that these storage fabrics are designed not just for capacity, but for predictable latency and continuous upgradeability, which is essential for ongoing research and production AI workloads. As the tour concludes, the video touches on retirement and repurposing strategies for old hardware, noting that retired DFMs can be repurposed to run virtual machines and other workloads while preserving data integrity through capacitors and parity-based recovery. The evergreen software model and modular hardware approach allow enterprises to migrate between generations with minimal or zero downtime, which is portrayed as a major value proposition. The host closes by situating Pure Storage within the broader context of modern data infrastructure, partnering with large-scale customers across cloud, AI, and analytics use cases, and inviting viewers to explore deployment options. Overall, the video blends technical exposition with a tour narrative to illustrate how high-end flash storage can deliver not only raw speed but also resilience, scalability, and flexible service models for complex data center environments.

Topics · technology · data storage · data center · enterprise it · hardware architectures · cloud & virtualization