Disaggregated routing is gaining traction as a way to reduce cost and increase flexibility as service providers face a major shift in the edge of their networks. The term is used broadly so it is useful to define disaggregated router:
A disaggregated router is an open approach to routing where the customer can choose hardware and software from a range of vendors that best meets their needs. This open approach enables innovation, avoids vendor lock-in, and drives cost down.
To dive deeper, a good starting point is to compare disaggregated routers to legacy routers. Using architectures from the 1990s, legacy routers are appliances where software is integrated with purpose-built proprietary hardware. This approach is monolithic where everything is tightly bundled together. It is a closed appliance which means the user has no choice in how to mix and match hardware and software. This limits the customer in terms of innovation to what the vendor has decided to support. Because these are closed systems, this results in vendor lock-in.
Disaggregation means separating the control plane and the data plane elements of the router. The control plane software determines the routes and runs on a centralized processing module that has a CPU and memory. This is both compute and memory intensive given the size of carrier routing tables. The data plane processing of packets occurs on IO modules that use ASICs to ensure line-rate forwarding. The cost of these devices is quite large which makes it costly to use them unless the port count is also high. Moreover, the control plane processing resources of CPU and memory do not grow as the port count increased.
Software Defined Networking (SDN) and open networking prompted a rethink of how to do routing. One of the key enabling technologies was commercially available switching ASICs from companies like Broadcom that were competitive with the proprietary ASICs made by the router vendors. In fact, some legacy vendors use these commercial off the shelf (COTS) ASICs in their hardware. These ASICs have matured to a point where they can even meet the demanding requirements of carrier networks. White box switches are open – meaning that a range of software can run on these devices. Increasingly, we are seeing carriers specifying their hardware requirements and the white box manufacturers have responded. This enables service providers to innovate with white box switches that are made for applications ranging from cell sites (DCSG) to core routers (DOR).
White box switches provide the data plane function of the router at a much lower price point than legacy routers. Customers have a choice of what software to run for the control plane. This can be delivered as a network operating system (NOS) which consists of an operating system (usually a Linux distribution) combined with the routing software including the various routing protocols like BGP, OSPF, and IS-IS. This is the most common commercial implementation. Another option is to use an OS like Open Network Linux and add your own routing software. This approach is used with the DANOS and SONiC open source projects which require more integration and have a different cost of ownership. In both models, implementations are standards-based so interoperability with most routers is broadly supported.
The NOS approach is still an appliance approach with a single routing instance and management domain. The cloud enables complete router disaggregation by running the bulk of the control plane in the cloud. This allows processing to scale far beyond the CPU and memory resources available on a white box switch. As a result, a single white box can support multiple virtual routers by running the control plane in the cloud while also ensuring full standards-based interoperability.