Carriers have dealt with metro environments for years. The geographical dispersion of users across large metropolitan areas has dictated how networks are architected and managed. Connectivity is provided across a number of edge links, and then those links are interconnected using a high-speed, high-capacity metro area network (MAN), typically using optical connections (think: DWDM). The objective is to pull traffic in, and then make distributed resources accessible via some distribution or transport layer.

While this is typical in carrier environments, can the same architectural principles be applied in the data center?

It turns out that the data center for hosting and *aaS players shares some of the same underlying requirements. Where carriers deal with connectivity to end users (businesses primarily), hosting providers have to service either tenants or applications. Regardless of the difference in end-user profiles, the need to provide connectivity and then transport traffic between resources is essentially the same.

Imagine a colocation environment with customers placing servers and storage within a data center. Those servers ultimately need high-speed access to Internet providers. The problem the hosting company has to solve is how to pull traffic from these servers and then provide transport between either other resources in the data center or to the carriers providing internet connectivity.

When all required resources and the links to the outside world are in close proximity (as with a single physical data center), the problem is actually fairly straightforward. Leaf-spine fabrics and MPLS gateways are sufficient in getting traffic between servers and then out of the data center. But what happens when those resources are not in close proximity?

For hosting environments that feature multiple data center sites, such an architecture is non-trivial. It requires careful engineering of transport traffic across the wide area network (WAN) between data centers, which makes turning up new servers and services complex. This creates additional operational overhead, which results in both time and money. Whether that burden is managed by the end user themselves or by the host, the time and cost are ultimately transferred to the end user.

But there are things we can learn from metro deployments.

Metro environments teach us that there are two underlying architectural requirements: end-user connectivity and high-speed transport. In the metro space, optical connections have been used for some time. Traffic comes in via access ports, and then it is transported via fiber (typically over DWDM equipment) that reaches across the metro area. This solves the performance issues that come with physically separate infrastructure.

It is well known that data center traffic is growing. The implication here is that data centers are becoming larger. At some point, physical constraints (power, cooling, and even real estate) mandate that these data centers get stretched over some distance. Additionally, latency-sensitive applications will tend to be placed closer to the edge, which also favors a distributed infrastructure. Finally, the emphasis on business continuity requires infrastructure to be spread out so that issues in one location do not take down the entire infrastructure.

All of these trends favor a shift in data center architecture toward something that more closely resembles metro deployments than legacy data center designs.

The cost of building high-capacity optical interconnects has historically been prohibitively expensive, but advances in manufacturing make some of the same optical capabilities that have powered large carrier networks accessible for the first time in data centers. The implications are profound.

A typical hosting service might offer VLAN-as-a-service to its customers. This allows individual tenants to buy VLANs that can be used to transport workloads across the data center. For modest requirements, a tenant might purchase one or a few VLANs. For more application-specific environments, those VLANs might be allocated to a single application.

With optical transport in the data center, the service model expands. Leveraging DWDM in the data center, it becomes possible to offer wavelengths as a service. Hosting providers can then house servers across the data center, using dedicated wavelengths to provide isolated, high-capacity, low-latency links between resource islands.

When those resources are distributed across multiple physical sites, the service is even more interesting. Customers can keep servers and storage in relatively close proximity to their end-users without stranding those resources. The optical interconnect between data center sites allows for transport between resources, effectively creating a single Layer 2 domain across an infrastructure that might span thousands of kilometers.

By stretching Layer 2 across a physical distance, applications can be turned up or even moved without touching the gateway connecting those sites. This gives customers the flexibility to manage their own applications and network services without going through a separate network team.

In some regards, deploying an optical interconnect provides a lot of the same functionality that exists in pure Ethernet transport today (through VPLS, for instance). However, DWDM adds another angle.

In some colocation environments, hosting providers essentially provide patch-panels-as-a-service. That is to say, they allow their tenants to pay for dedicated cross-connects to provide low-latency transport between resources. In a DWDM environment using a wavelength-as-a-service model, hosting providers can improve on these types of offerings. Programmable optical fabrics allow providers to allocate bandwidth on demand, effectively giving their tenants a way to dial bandwidth up (and down) between resources (or sites) as needed. This changes the pricing model from fixed to more pay-as-you-go, which results in a more flexible and ultimately less expensive service to end users.

Finally, while these capabilities lead to new products for co-location providers, they can be applied within any distributed data center environment. The use of wavelengths to dynamically create high-speed, isolated paths between resource islands allows for a more dynamic data center fabric that can extend well beyond the physical limits of more legacy architectures.

Metro carriers have successfully used this technology for years. It’s now possible for others to take similar advantage without breaking the bank.