It is rare when the future of virtually anything is completely divorced from its present. But that is the case with enterprise data centers. Looking at how mainstream enterprise data centers will evolve—both in the near and medium term—it’s important to understand that progress will be evolutionary. Despite every technologist’s preference, pragmatism dictates that our ties to the past will not be completely severed.
This means that the architectural building blocks responsible for data centers of the future need to be effective at handling both legacy and modern applications, providing a bridge from where enterprises find themselves today to where they need to be tomorrow.
Requirements for the Data Center
First and foremost, it must be acknowledged that most data centers will host legacy applications for the foreseeable future. Despite widespread adoption of cloud-delivered applications, the reality is that there is a long trail of legacy applications that are either not well-suited to the cloud or there is no business justification to modernize. So despite the best of intentions, most mainstream enterprises will find it impossible to completely separate from their past.
Of course, some applications will be rewritten while others will be added. Any application developed today will likely leverage microservices architectures and be built on common application platforms such as Kubernetes and CloudFoundry. For applications hosted in the cloud, the data center architecture is left to the cloud providers, but not all workloads will run in the public cloud. For some applications, infrastructure will be on-premises (as with multi-access edge computing required for distributed IoT workloads, for example). In these instances, the data center is somewhat more nebulously defined, not only for private cloud but also hyperconverged infrastructure running closer to the user. This means the data center of the future must account for diverse deployment models leveraging these application platforms.
Finally, these applications will largely be consumed on-demand. This places the operational burden on the architectures, as user demand will dictate when and where connectivity will be required. As applications become more distributed, data centers will need to support a self-service ability to create connectivity services between endpoints within the data center, which will be a combination of on-premises infrastructure and one or more public clouds.
Operations as the Unifying Problem Space
The unifying problem space in the future of data center networking is operations. Whether it’s providing a common operating model for both legacy and modern applications or supporting self-service connectivity services across distributed data centers, the key is converging on a narrow set of operational tools and protocols that serve as an abstraction layer for the underlying infrastructure.
Without a simpler, more efficient and highly-repeatable operating model, enterprises will find that they are incapable of satisfying the disparate needs of the future without compromising on lead times, integrated security and mean time to repair.
EVPN enables this operating model. It allows data center operators to create virtualized L2 and L3 networks that meet the needs of both legacy and modern applications directly on top of the IP underlay, without introducing the additional complexity of MPLS (which was required in the past).
Supporting the Past
For most enterprises, the majority of legacy applications will require L2 connectivity. Where data centers of the past used to feature Spanning Tree, the current best practice for building scalable, highly-available data center underlays is to use L3 protocols. In modern data centers, L2 connectivity services for legacy applications must be virtualized on top of L3 IP connectivity. EVPN can use VXLAN to accomplish this.
EVPN provides a significant step forward, even in L2 environments. There are no forwarding loops in EVPN L2 domains, which greatly reduces the risk of outage due to human error. Additionally, EVPN features a more efficient handling of broadcast and unknown traffic, helping scale VPNs with legacy applications deployed in large L2 domains. EVPN also supports rich multihoming capabilities, allowing applications to maintain multiple connections to their virtual network segments—useful for improving resiliency and distributing application load across the network.
All of this is provided while maintaining strict tenant separation. Each application is connected via its own virtual segments and traffic is passed based on separate edge policy. This creates a separation of dependencies as each application is encapsulated into its own VPN, granting fine-grained control.
So not only does EVPN support legacy applications, it does so in an inherently more scalable and operationally friendly way.
Supporting the Future
For modern applications, the future involves microservices leveraging containerized application components orchestrated via application platforms like Kubernetes. Distributing application components creates a dependency on standardized connectivity services that can extend to containers, VMs and bare-metal servers deployed across both private and public clouds.
In these environments, connectivity services extend beyond merely passing packets. If there is a routed instance that acts as the virtual network edge, that virtual device can also serve as a point of policy enforcement. By building policy and control into these virtual network endpoints, connectivity services can be easily extended to include security (i.e., microsegmentation).
Of course, modern applications are not necessarily static. One benefit of microservices is that they can scale up and down dynamically, effectively right-sizing infrastructure consumption based on demand. This places a real-time operational burden on connectivity. Combining EVPN’s control plane with an over-the-top orchestration solution, for example, can provide the dynamism required to service a modern application.
Diverse Environments, Common Operations
If the future of data centers requires maintaining one foot in the past and the other stepping toward the future, technology building blocks will be required that support both. The key is doing so without adding to the already burdensome operational load for most enterprises. EVPN’s support for virtualized Layer 2 Ethernet and Layer 3 IP services, with first-class connectivity multihoming, robust multicast features and optimized handling of “network noise,” makes it a natural choice for forward-looking data center architectures focusing on simplified networking.
Today, we kick off the third and final video in a series of content designed not to sell but rather to illuminate. With our GetSmart series on SDxCentral , we will look at why EVPN matters, providing an industry view of what EVPN can do as the de facto architectural foundation for the modern data center. For additional information please download our E-Book on EVPN-VXLAN .