Jason talked with Thomas Pfenning, General Manager of Windows Server and Cloud Division, and Zhang Lei, Director of System Architecture, Alibaba Group, about rack-scale architecture, an evolution of rack designs that enables independent upgrades of compute, network and storage subsystems, and improves rack utilization and efficiency.
Maximize flexibility, minimize complexity and costs
The cloud executives talked about the need to have an infrastructure that was flexible enough to support a variety of workloads, without the need for customized hardware. “In Microsoft’s datacenters, where we host a variety of workloads (such as Bing, Xbox Live and Office 365), the most expensive infrastructure is unused infrastructure,” stated Thomas. “The key to avoiding unused infrastructure is using the same repeatable building blocks to host as many of your workloads.” To that end, rack-scale architecture could increase the use of more standardized hardware resources with enhanced flexibility to run a variety of workloads.
Next steps with rack-scale architecture
Zhang Lei commented that Alibaba has teamed with Tencent and Baidu around an initiative for Chinese providers called “Project Scorpio” to define reference architectures around shared cooling and power. Alibaba also intends to pilot Intel Silicon Photonics Technology, a high bandwidth interconnect for rack scale architectures.
Thomas mentioned, “We are seeing more change to our datacenter architecture in the next 5 years, than in the last 20.” With these changes, he acknowledged that the operating system needs to work at a rack-level, instead of just at a computer level.
Watch the conversation between the cloud executives here.