The world is moving to hyperscale—it’s time to rethink the datacenter
I was thrilled to present at the Datacenter Dynamics (DCD) Enterprise summit today. One of the conference themes was “Is the datacenter dead?” From where I sit, it certainly is not, but we are in a period of unprecedented transformation.
The vast majority of workloads over the next five years will be deployed in hyperscale datacenters—either you will operate a datacenter at scale, or you will operate within someone else’s. Cloud computing as a deployment model, whether public cloud or on-prem, is driving new thinking around the design of the datacenter. Enterprises with their own datacenters will operate at a bigger scale, and continue to adopt best known practices from cloud service providers.
At scale, efficiency is critical. The principles of open standards, manageability, and flexibility are foundational to the next generation of datacenter architectures. Efficiency of the datacenter will be driven by optimized software defined infrastructure (SDI) and the move to Rack Scale Design.
Optimized software-defined infrastructure efficiently places workloads and composes resources
Software-defined infrastructure (SDI) performs two critical roles. First, it needs to efficiently manage and compose resources across compute, network, and storage. Second, it exposes critical hardware functions to an application.
The quest to improve SDI orchestration means a shift from creating virtual partitions of resources to shifting toward composition of resources to meet application requirements. Standard management APIs are required to enable a range of different SDI solutions, such as those by VMware*, Microsoft*, and Red Hat*, with the ability to support bare metal, virtual machines (VMs), function-as-a-service (FaaS). With the rise of artificial intelligence (AI), SDI increasingly requires the ability to run a broader range of applications across the environment supporting both homogeneous and heterogeneous infrastructure. To make the most of SDI, the underlying infrastructure also needs to evolve.
Rack Scale Design, a pooled & disaggregated architecture optimized for SDI at scale
The datacenter building block is changing. In order to be efficiently utilized, an infrastructure based on resource pooling and disaggregation is better matched to the balance of workloads required. Intel’s goal with Intel® Rack Scale Design (Intel® RSD) is to enable pooling across datacenter components, including compute, storage, network, and management. As design points shift because workloads are rapidly evolving, the disaggregation of RSD enables late binding of resources and technologies. This means that hardware resource investments can be scheduled for when the workload requirement is understood.
Intel is investing heavily to enable the shift toward RSD and datacenters that are optimized for hyperscale. I’m happy to announce, as a tech preview, that we are ready to demonstrate with our partners the technical capabilities and use cases provided by latest pooled FPGA technology enabled through RSD APIs.
Learn more about Intel® Rack Scale Design here.