Revamping a Global Network Architecture

Cloud fabric network

With its software subscriptions and revenue growing more than 20 percent year over year, Workday had to address its legacy technology infrastructure. The company's data centers were running out of capacity, and its network was unable to scale and adapt to customer and business needs that continue to expand and evolve.

"We needed an entirely new network architecture," says Ahsan Ghayas, senior principal engineer and lead architect for Workday.

The company began working on a design for a "cloud fabric network" that would connect five new, globally dispersed data centers and support Workday's next phase of business growth. Horizontal scalability, high capacity and availability, modular oversubscription, and full automation were the primary requirements. Ghayas just needed to find the network hardware and software that could deliver.

"We had a great experience with Cisco Nexus switches in our legacy data centers, and it became clear Cisco Nexus 9000 Series Switches were the best choice for our cloud fabric network," he says. "They give us exceptional density, capacity, and scalability as well as highly desirable line rate encryption capabilities."

One model, many views

With design validation, software guidance, and operational best practices from Cisco CX, Workday deployed a 100G network with five separate fabrics—one each for three new data centers in the United States as well as new computing sites in Ireland and the Netherlands. The network currently has 8000 Cisco Nexus 9000 Series switches but can scale to 15,000 switches as Workday's business grows. All of the network fabrics are centrally deployed and managed with open system tools and Cisco NX-OS.

"NX-OS is very mature, it offers high availability features like stateful switchover, and it has open APIs that make it easy to integrate with tools like Ansible and Python," Ghayas says. "We used those tools to automate the entire fabric rollout for each data center."

Doing so accelerated the deployment of full compute pods—which include 48 server racks and two spine switches—from six days or more to less than a day. And those deployments were completed amid a pandemic-related lockdown with significant supply chain challenges.

"Each data center has 1600 switches, 1200 racks, and 36,000 server nodes," Ghayas explains. "With only a dozen people deploying it all, automation was critical."

It's been just as important after deployment, he adds, enabling Workday's lean network operations team to manage a much larger environment without additional staff members.

"We have a 'one model, many views' philosophy," Ghayas says. "We want to use the same model and tools not just for network design and deployment, but also for ongoing observability and monitoring as well as policy creation and governance."

Workload portability

In addition to being linked to each other via wide area network (WAN), Workday's new data centers are also tightly connected to public cloud environments.

"We have a tremendous amount of flexibility for customer deployments and seamless workload portability—across regions, data centers, and clouds," Ghayas says.

With Workday's new data centers fully operational, Ghayas has his sights set on additional automation and more observability across the company's network fabrics and compute platforms.

"We'll continue to improve our visibility, speed, and efficiency," he says. "And we'll eventually make the leap to 200G and 400G with Cisco Nexus 9400 Series switches."