This blog was co-authored by Jeff Schneider, Senior Consultant, Product Management
The latest generation of PowerMax 2500/8500 models are the first mission-critical storage systems to integrate the NVIDIA BlueField DPU (data processing unit) technology into the architecture. This milestone is a testament to the long-standing relationship between Dell Technologies and NVIDIA. Let us explain.
Many leading businesses and organizations around the world rely on mission-critical, enterprise storage arrays to provide super-high availability and predictable high performance. Medical databases, financial transactions, airline operations systems, energy system controllers – pick your favorite example. Some things just have to work, reliably, all of the time! The businesses’ success depends on it. In some cases, people’s lives depend on it. It takes focused attention to the system architecture design to achieve this. Performance matters, scalability matters, resiliency matters, power and space efficiency matters, and cost of ownership matters. This mantra and a deep obsession with performance, scalability, reliability, and efficiency is the reason why Dell’s PowerMax is the leader in this space today.
The Dell PowerMax platform is built on decades of experience and innovation. The integrity and constant availability requirement of customer data, with predictable high performance, and security built in from the ground up are key factors in the design of the system architecture. Add to that the ability to expand capacity and compute power in a flexible and cost-effective way to meet customers’ changing needs. The PowerMax system architecture and PowerMaxOS 10 software platform provide a unique balance of building on a robust, mature base while incorporating new innovative features and the latest available technology across the industry. This is where our work with NVIDIA BlueField DPU technology comes into play.
“For Salesforce, trust is our number one value, and PowerMax has been a key part of that in terms of availability, reliability, and performance.” – Pete Robinson, Director of Infrastructure Engineering, @Salesforce
Our collaboration with NVIDIA on optimizing storage SANs has spanned several years leading up to our integration of NVIDIA BlueField DPU technology into the PowerMax architecture — specifically, into the dynamic media enclosures (DMEs) developed for PowerMax.
The new DMEs are not just another conventional drive enclosure. These are “smart,” fabric-attached units that each house up to 48 NVMe SSDs. The enclosures have dual LCCs (link controller cards) for high availability. Each LCC includes a BlueField DPU, a multicore system on a chip (SoC) from NVIDIA. The core of the PowerMax architecture is built around NVMe-over-Fabrics with NVIDIA Quantum InfiniBand as the transport layer. The BlueField DPUs are key to making this possible.
The benefits of this architecture are huge. In this latest generation of PowerMax, the compute nodes and the dynamic media enclosures are all connected on the dynamic NVMe/InfiniBand fabric. This allows for the compute and media elements to scale independently. Have an application that is highly compute-intensive? Add more compute nodes. Need more capacity? Add more DMEs and NVMe flash drives. The PowerMax 8500 can scale out to 16 nodes and up to 8 DMEs, maxing out at over 18PB of effective capacity.
Another benefit of this dynamic fabric architecture is the fact that any node on the fabric can access any data drive in the system, regardless of which DME it’s physically located in. Customers see this benefit directly in the performance they achieve with the array. I/O from any host connected to any node in the array can be routed to any drive, with low latency and high efficiency. This new, DPU-enhanced architecture streamlines this access. No extra “hops” are required through adjacent nodes to get to the data.
Any-to-any access aids with system resiliency as well (some things have to work all of the time!) since access to any storage is not dependent on any specific node or data path. If, as could happen with any electronic device, there is a failure of some kind that causes a node to go off-line, no data is stranded and no extra overhead is inserted as other nodes pick up the slack. This kind of data availability is critical to the users of PowerMax.
The working relationship between Dell (and EMC prior to the acquisition) and NVIDIA has spanned multiple product generations and has yielded significant advantages to the PowerMax platform. Multiple NVIDIA components are used in the system, in addition to the BlueField DPUs in the DMEs. The ConnectX smart adapter technology is used on the initiator side of the InfiniBand fabric as well, and NVIDIA switches are used in the larger 8500 system. The BlueField DPUs run custom code developed in-house at Dell, allowing direct access to all of the features on the chip.
PowerMax has leveraged T10-DIF technology to enhance data integrity for multiple generations. The NVIDIA components provide hardware offload for this functionality, allowing for maximum performance with the protection that T10-DIF offers. Any workload that can be offloaded from the main compute module CPUs helps with overall array performance. The Arm cores in the BlueField DPUs are utilized for low-level drive management, background scans, hot plug management, and other functions. The entire system is optimized for performance.
The BlueField DPUs also support secure boot, further enhancing the security of the PowerMax platform using built-in hardware acceleration. PowerMax is designed for Zero Trust security architectures1, with end-to-end security features in place to safeguard customer data. It’s another critical aspect of being the leader in mission-critical storage.
“PowerMax is the cornerstone of the data center. The PowerMax 2500/8500 cyber resiliency capabilities and intrinsic value of data safety give us peace of mind for trusting enterprise applications on PowerMax.” – John Lochausen, Technology Architect, World Wide Technology
The dynamic InfiniBand fabric, enabled by the BlueField DPU and other NVIDIA technology, like NVMe-oF Target Offload, is key to the advantages of the PowerMax architecture over conventional storage arrays. When combined with the other features of the architecture, such as a 7x increase in capacity per node2, a 14x increase in capacity per rack unit3, an increase in guaranteed data reduction to 4:14, and a 64-bit file storage capability, you can see why PowerMax is the leader in mission-critical enterprise storage.
1 Based on Dell’s internal analysis of cybersecurity capabilities of Dell PowerMax compared to Dell’s seven pillars of Zero Trust architectures, March ‘22.
2 Based on Dell’s internal analysis comparing Effective Storage Capacity of the PowerMax 2500 compared with the PowerMax 2000, March 2022. Actual storage capacities will vary.
3 Based on Dell’s internal analysis comparing Effective Storage Capacity per rack unit (1.34”) of the PowerMax 2500 compared with the PowerMax 2000, March 2022. Actual storage capacities will vary.
4 Based on Dell‘s Future-Proof program that offers 4:1 data reduction guarantee based on PowerMax data reduction tools (dedupe and data compression) for open systems storage, March 2022. Actual data reduction rates will vary.