The primary objective for networking for decades has been to provide connectivity between resources. In the early days, mere connectivity was valuable enough to make it a service by itself. Then the game shifted to providing bandwidth – essentially more connectivity. Since that time, we have tried different variations of the connectivity-as-a-service model: clean pipes for more secure connectivity, patch panel services in the datacenter for faster connectivity, and VLAN-as-a-service for more isolated connectivity.
As connectivity becomes more and more a commodity service and requirements become more demanding, how that connectivity is provided has to evolve.
In the datacenter, the network serves as an interconnect between pools of resources. For the most part, those resources exist in close proximity (within the same physical datacenter, and typically within a narrow space within that datacenter). This is because the act of separating resources over even relatively short distances incurs performance hits as traffic gets shunted through aggregation and transport layers.
There are technologies that make transport higher capacity and lower latency, but these have typically been relegated to large carrier networks primarily because of cost and secondarily because of the specialized nature of Layer-1 equipment. But as manufacturing processes improve, the cost comes down, and the introduction of technologies like SDN helps remove the complexity.
The result is a new set of capabilities in the datacenter that can both improve performance and enable new services.
Imagine the datacenter as having two jobs: get traffic from resources (servers, storage, etc.) and move that traffic to other resources (other servers, storage, or up and out of the datacenter). Using this model, there are basically two classes of connectivity service: physical connectivity (paying for dedicated bandwidth or links) and virtual connectivity (VLANs, for example).
When you introduce optical transport into the datacenter, you essentially create a different way of offering connectivity service to datacenter tenants. Using DWDM as an interconnect between resources (be they in a single datacenter or across large distances), providers can offer a wavelength as a service.
The wavelength-as-a-service model is a bit of a hybrid between the bandwidth and VLAN models. It allows users to get guaranteed bandwidth across an isolated connection. You essentially get a single Layer-2 domain stretched between resources that are physically distributed.
Combining this with a central control model typical in SDN architectures makes the offering even more compelling.
[To see how DWDM is changing the way multi-site datacenter fabrics work, check out Iron Mountain and Plexxi on SDNCentral’s DemoFriday on July 18 at 1p EDT. Register here.]
Optical interconnects are interesting because they allow single-hop connectivity across multi-hop topographies. Put differently, an optical path can traverse multiple devices without incurring a switch hop in between. Traffic is photonicially switched without hitting the switching ASIC for a lookup. This allows for very low-latency end-to-end transport via direct connections. Imagine if every flight in the US had to travel through Chicago. Direct connections affect experience.
Perhaps more importantly, those optical paths can be programmed. They are not statically bound to the physical devices. This means that paths can be set up and torn down dynamically. Imagine that an application requires the exchange of a lot of data (as with a Big Data app). An optical path can be established between resources, allowing a high-capacity, single-hop connection to be established on demand. This helps guarantee connectivity and performance while simultaneously isolating the workload from the rest of the network (effectively protecting both the application and any residual traffic on the network).
From a provider point of view, this represents a different type of connectivity service. The ability to add bandwidth on demand anywhere in the network eliminates the need for expensive, dedicated connectivity. Customers can subscribe to baseline connectivity services, and then use additional wavelengths when and where necessary. This creates a service model that more closely resembles the cloud pay-as-you-go services that are attracting more and more attention.
More interestingly, these optical paths are not limited to a single physical datacenter. Optical transport can easily handle metro areas, and without additional equipment span more than 1,000km. This enables providers to create connectivity services across multiple datacenter sites, which allows customers to locate their servers wherever makes sense based on proximity, business continuity, and even physical (space and power costs, for example) requirements.
While there will be additional hosting and provider services that emerge as a result of technologies like SDN and NFV, it could be that the expansion of existing connectivity services provides the best short-term opportunity for manageable changes in product and service offerings within the hosting and CoLo space.[Today’s fun fact: The average lead pencil can draw a line 35 miles long. That’s 50,000 English words according to the guys at Snapple. Who knew?]