Whether you’re starting out on a fresh playing field or diving into a mud pool of decades-old complexity, designing and deploying a new or modernized data center is a rewarding endeavor; not just for the engineers and architects, but also for the businesses that reap the benefits of agility, scalability, and performance that come along with it.
And the first step on that road is to talk. The initial conversations with thought leaders, business strategists, and technical architects are the most pivotal in the discovery phase of any large project. It is at this phase that the box is forming, and questions must be asked outside of it to shape its dimensions. To transform the network, you must be prepared to ask challenging questions that drive conversations around open networking, automation, modularity, scalability, segmentation and re-usability. Before vendor selection, it is essential to compile a list of business and technical requirements founded upon a set of guiding principles.
Here are seven to keep in your pocket:
1. The network architecture should use standards-based protocols and services
2. The network should be serviceable without downtime
3. The network architecture should promote automation
4. The network should be consumable
5. Physical boundaries should not restrict segmentation capability
6. The network must be scalable
7. Network changes should be verifiable before implementing in production
Let’s quickly define each of those principles.
The Network Architecture Should Leverage Standards-Based Protocols and Services
Proprietary protocols and closed ecosystems require highly specialized engineers, limit inter-operability, and force organizations into particular designs that are difficult to escape. Where possible, standards-based protocols should be leveraged, and open ecosystems embraced which promote interoperability and innovation.
With that in mind, Cumulus Networks has pioneered standards-based protocols like Ethernet VPN (EVPN), an extension of BGP that serves as a VXLAN control plane, simplifying the virtual stretching of tenant domains across massive physical networks while still offering interoperability with other standards-based implementations of EVPN.
The Network Should Be Serviceable Without Downtime
It should go without saying that fault tolerance is a necessity. All compute nodes must be dual-connected to redundant upstream Leaf switches. Leaf switches should have redundant peer-link connections between each other, and to each Spine switch. Equal-cost multi-pathing ensures that all paths are active and forwarding. Inserting or removing a Leaf or Spine switch should not affect production traffic. Scripted BGP traffic engineering techniques executed from Cumulus Linux allow administrators to automate hitless patching and upgrades.
The Network Architecture Should Promote Automation
Manual configuration changes are time-consuming and prone to human error. Validating that the network is running as intended and adheres to network and security policies is too often overlooked when designing or monitoring a network. Automating tasks can make the network self-healing, more consumable, and easier to audit. Familiar Linux APIs, such as those leveraged in Cumulus Linux, allows DevOps engineers to integrate the network into automation engines without the friction of dealing with numerous, vendor-specific APIs. Having the same network operating system (NOS) on each device, regardless of the underlying hardware, opens the door for simplified network automation.
The Network Should Be Consumable
Tied into automation is the concept of consumable self-service networks. Whether the data center is private and serving a single organization, or built for a busy IaaS platform, having the capability to empower administrators or customers with self-deployable networks should be a key consideration with new network designs. Creating networks in the public cloud is a fundamental feature everyone expects.
Customers should have the capability to deploy segmented networks on the fly, without the intervention of network engineers. A Linux NOS like Cumulus Linux is ideal for orchestration solutions, due to native Linux modules and APIs. Common deployments of Cumulus Linux harness EVPN with automation, freeing network engineers from the tedious deployment of new networks while simultaneously enabling customers to build their own on the fly.
Physical Boundaries Should Not Restrict Segmentation Capabilities
EVPN is used to compartmentalize and segment tenant traffic across the data center environment, providing an open and flexible architecture irrespective of physical boundaries, transporting network segments anywhere in the data center or across data centers. Enhancements like BGP unnumbered simplify the automation of EVPN fabrics with Cumulus Linux. Modular portability is critical when thinking about network design.
The Network Must Be Scalable
A Leaf-Spine Clos architecture is ideal for data centers; with equal-cost multipathing of 128 links, Leaf-Spine pods can become massive. Additional pods can be added to grow horizontally, or new tiers to grow vertically, interconnecting indefinite numbers of pods. EVPN scales with the physical topology, providing the ultimate modularity for scale. If port-density or port-speeds in specific areas become insufficient, Cumulus Networks’ disaggregated model allows data center admins to swap hardware modularly, automating the NOS and network provisioning with ONIE, proving flexibility at the micro and macro scale.
Network Changes Should Be Verifiably Testable Before Implementing Into Production
Downtime and SLA violations can cost organizations significant dollars in the form of refunds or reputation. With Cumulus VX, organizations can reduce the risk of downtime by fully simulating network changes and upgrades before flipping the switch and making them live. The exact software and capabilities that run Cumulus Linux also run Cumulus VX, assuring that simulated tested network changes will be successful on systems in production.
When designing your next data center network, carry these guiding principles with you from project inception through to network deployment. While the list is far from all-encompassing, these ideas will help generate specific results for a highly effective and agile data center, built to scale, and designed to lead.