Providing Innovative Solutions to Hyperscale Data Center Challenges

Some of the challenges of today’s data center infrastructure include underutilization of resources, inefficiencies in IT operations, and expensive refresh and scale-out options. The typical utilization rate for server CPUs are under 50 percent, even with virtualization. Intel® Rack Scale Design (Intel® RSD) addresses these challenges with an innovative architectural design that replaces traditional servers with disaggregated hardware “pools” of compute, storage, and accelerator resources.

These pooled resources are composable; they can be stitched together at the software level and exposed at the orchestration level as virtual or physical servers. All of this is based on industry standard APIs to allow for interoperability between vendors. The benefits of this design include maximized resource utilization, increased agility, and lower costs. In Intel’s newest version of RSD, we have introduced our newest feature of pooling NVMe storage over fabrics using Ethernet.

Intel® RSD Release 2.3 provides new capabilities for storage pooling for Supermicro Rack Scale Design (SRSD)

Supermicro, a long-time collaborator with Intel, has built the Supermicro Rack Scale Design (SRSD)* product line based on the Intel® RSD specifications, and continues to evolve the line with each new version of Intel® RSD. With SRSD 2.1, customers were able to disaggregate their SAS/SATA storage using iSCSI (Internet Small Computer Systems Interface), improving storage utilization by allowing customers to assign storage where it is needed most.

With SRSD release 2.2 Supermicro improved upon this pooled storage option and allowed for disaggregated NVMe storage over PCIe via point-to-point connections between compute servers and storage. Storage pooling over PCIe provides very high bandwidth with low latency due to the direct speed from the PCIe bus. However, PCIe is inherently limited by cable length which in turn limits the ability to span across multiple racks and the total amount of storage available for composition.

This limitation is addressed in Intel® RSD v2.3, which Supermicro is now incorporating into the SRSD product line. With this latest version, disaggregated NVMe storage can be connected and pooled over fabric, using RDMA-enabled network interface controllers (NICs) to connect storage to servers through Ethernet cards (Figure 1). Pooling storage in this way overcomes the scalability and cable-length limitations present in NMVe over PCIe, improving scalability and connectivity across multiple racks while maintaining high bandwidth and low latency. Assuming a 33-percent utilization rate generally seen in today’s data centers, Supermicro estimates a company can realize an approximately 45-percent savings in CapEx with a disaggregated architecture of eight nodes sharing 32 drives. With a disaggregated architecture and composable nodes, a company could run even more workloads in parallel, resulting in an increased utilization of these existing resources.

hyperscale data center
Fig. 1: A visualization of NVMe over Fabrics using RDMA Ethernet in an Intel® RSD-based solution which provides increased scalability across multiple racks inside a POD.

Storage pooling is fundamental to Storage as a Service

Service providers can offer a dedicated storage environment where storage is rented by other companies or individuals. The storage provider then supplies the client with the software required to transfer, access, and back-up their data, among other specific workloads. This service is sometimes called Storage as a Service (STaaS).

STaaS provides elastic orchestration including the ability to move data intact between systems when necessary as well as attach and detach resources easily. However, service providers suffer from the same poor utilization rates. In order for service providers to maximize the utilization and efficiency of their hardware infrastructure, dynamic allocation of this storage is key.

The newest release of Intel® RSD provides this key ability to pool storage and integrate storage management via industry standard APIs which improve the delivery of STaaS substantially. By deploying a software-defined infrastructure on Intel® RSD 2.3, service providers can benefit from improved operational efficiency and utilization, shorter time to market, and a lower total cost of ownership. Ultimately, Intel® RSD could improve the profitability of cloud-based storage services.

The newest RSD 2.3 benefits many kinds of data centers

Intel® RSD provides disaggregation, composability, and interoperability with added improvements in storage pooling over fabrics and STaaS management. Supermicro has taken advantage of these capabilities from Intel® RSD in their new SRSD 2.3 product. Cloud service providers, communication service providers, and enterprises alike can take advantage of these improvements. They could see enhanced utilization of resources, efficiencies in operations, and substantial savings.

To learn more about the benefits of Intel® Rack Scale Design, visit intel.com/intelrsd.

Supermicro Rack Scale Design* is available for purchase now. Visit the Supermicro website to learn more here.

Written by Chris Buerger, Head of the Customer Success team for Intel Rack Scale Design at Intel and Arun Kalluri, Senior Director of Rack Scale Solutions & Systems Management Software at Supermicro.