Edelweiss - Fotolia
What is Intel Rack Scale Design and how does it work?
Find out what Intel's Rack Scale Design architecture is and how its reference code and APIs enable Intel-partner products to compose on-demand virtual servers for specific workloads.
Intel Rack Scale Design is a reference architecture, or blueprint, for building a composable disaggregated infrastructure that utilizes software-defined technologies to deliver a hyperscale data center environment. The infrastructure is made up of rack-based hardware resources and software components based on Intel's Rack Scale Design reference code and API specifications, which make it possible to compose operating environments on demand to support specific workloads.
Introducing Intel Rack Scale Design
The Intel Rack Scale Design infrastructure separates the physical compute, storage and network resources within a rack into logical resources that administrators can access through management APIs, which provide interfaces for discovering, composing and monitoring resources. The infrastructure employs a high-speed interconnect fabric that links the disaggregated resources and makes it possible to compose them into virtual servers within seconds.
Intel Rack Scale Design assembles the logical components into resource pools accessible to the APIs to compose nodes. A node is a grouping of logical system elements -- such as CPUs, memory modules, storage components or network switches -- optimized for specific workloads.
For example, a workload might require a node that includes a four-socket compute module, a 64 GB memory module and an 8 TB solid-state drive. Once the node has been created, the workload can use its resources to carry out operations. When the workload no longer needs the node, its components are released back to the resource pools and made available for other workloads.
Rack Scale Design can help save on hardware costs because IT teams buy only the equipment they need when they need it. The composable infrastructure can also increase resource utilization, and it can reduce administrative overhead because many operations are automated and components are faster and easier to refresh and scale out.
Because Intel Rack Scale Design is based on industry standards, organizations can choose the hardware components that best fit their circumstances and budgets, which can, ideally, help them avoid vendor lock-in and the costs that sometimes come with it. The composable architecture also offers more flexibility for supporting today's fluctuating and varying workloads.
Intel Rack Scale Design components
An Intel Rack Scale Design environment is made up of both hardware and software components that together create a composable infrastructure that can span multiple racks to deliver a hyperscale data center. The following figure provides an overview of a typical Intel Rack Scale Design implementation.
At the heart of the infrastructure is the rack and its components. Each rack is made up of multiple drawers that house the physical resources. A drawer normally contains a blade, module or other element, which, in turn, contains the compute, storage or network components. The exact approach to housing the hardware can vary, but conceptually, the resources are generally presented in this way.
Intel's Rack Scale Design infrastructure utilizes a software component called the Pooled System Management Engine (PSME) to interface with the physical resources. The PSME provides information about the availability of resources, and it exposes a set of APIs to higher levels in the management stack. In a common Intel Rack Scale Design configuration, each drawer is associated with a PSME instance -- although the PSME can also run at the blade, module or rack level. The Intel Rack Scale Design vendor partner, which sells products using the composable architecture, determines how the PSME is implemented.
Each rack also includes a software component known as the Rack Manager Module (RMM), which communicates with the PSMEs and handles rack-level functions, such as managing power, reporting environmental status or distributing security certificates to the PSMEs. As with the PSMEs, vendors determine how the RMM is implemented.
In a typical Intel Rack Scale Design configuration, each rack also supports both a data network and a management network. The networks are based on networking technologies such as 100 Gigabit Ethernet, InfiniBand or optical interconnects. In addition, the rack includes a top-of-rack switch that facilitates communications with other systems.
The top-of-rack switch also connects racks into a pod, a common management framework for scaling the infrastructure beyond a single rack. Integral to this framework is the Pod Manager (PODM), a software component for managing functions within and across the racks that make up the pod. For example, the PODM can power-on modules, initiate provisioning, discover resources or compose nodes using resources from any of the racks.
The PODM also communicates with orchestration software, which can request the PODM to carry out management operations. The orchestration software sits outside of the Rack Scale Design infrastructure, utilizing the APIs exposed by the PODM to perform and automate operations -- most notably composing infrastructure nodes on demand to meet the needs of specific workloads.
All Intel Rack Scale Design-related communications on the management network rely on the APIs exposed by the PSME, RMM and PODM software components to carry out operations. Intel based the APIs on the Redfish RESTful framework, an open standard for data center management. The Distributed Management Task Force (DMTF), a group of organizations that create open manageability standards for new and traditional IT infrastructures, defined the Redfish framework.
Today's Intel Rack Scale Design implementations
A number of vendors now offer composable products based on the Intel's Rack Scale Design architecture, including American Megatrends, Canonical, Dell EMC, Inspur, Supermicro and several others, with each vendor putting its own spin on the infrastructure. For example, the Supermicro Rack Scale Design offering includes the PODM, RMM and PSME software components, as well as a pooled NVMe controller and web-based user interface. The software is part of a complete Intel Rack Scale Design infrastructure product built from Supermicro server, storage and switch hardware.
Organizations considering Intel's Rack Scale Design infrastructure should carefully review and compare different products based on the composable reference architecture before making any decisions. Intel gives vendors a great deal of latitude when building Rack Scale Design systems, so products can vary widely. At the same time, keep in mind that it is a relatively young technology. Although Intel designed the infrastructure to be hardware-agnostic, options remain limited, and so the potential for vendor lock-in continues in spite of the reference architecture's overall potential to avoid such a scenario.
Intel Rack Scale Design has been gaining steady traction, however, which will likely lead to more options and advancements in the technology. This will benefit enterprises ready to embrace the composable infrastructure reference platform going forward.