Disaggregating the Appliance Model

Disaggregating the Appliance Model

In carrier networks, the appliance model is giving way to a disaggregated model. Traditionally, functions like routing were implemented in a tightly coupled hardware and software bundle. Routing vendors could ensure the proper operation of their system because they controlled the entire network element. They even developed their own switching ASICs for their line cards. Combined with the complexity of the software, this created significant barriers to entry.

Service provider-oriented routers evolved to scale up to meet the demand for more bandwidth and higher speed ports at the edge. The router chassis got bigger to hold more line cards and new line cards supported more higher speed ports. This had the effect of keeping the price per port down provided that the chassis was reasonably full. However, too many empty slots worked against this since the cost of the common components (chassis, power supply, and routing engine) was spread over a small number of ports. The marginal cost per port could be astronomical when the chassis was full, and you needed a new chassis for just a few more ports.

The economics of service provider networks are changing rapidly. In the new service edge, IP will be pushed all the way to the cell tower with technologies like eCPRI. Crosshaul is critical to enabling the disaggregation of the RAN components as the BBU is implemented as a VNF way form the cell site. We had addressed this trend in our recent webinar with IHS. Once IP moves in, routing follows. As the location that needs routing gets closer to the customer, there will be a massive increase in the number of locations and those routers must get much smaller. We expect to see an increase of two orders of magnitude in locations. It is also clear that the cost must come down by a similar order of magnitude.

Service providers are already embracing disaggregation as a path to scale out the service edge while driving down cost. TIP’s DCSG project specified the lower scale requirements for routers at the small cell sites. It is important to note that they also want carrier grade protocols for this application.
The benefits of router disaggregation are not limited to this very edge of the network. Crosshaul locations (BBU hotels) and PE routers also benefit from router disaggregation. However, unless router disaggregation breaks the appliance model, these benefits apply only to CapEx.

Router disaggregation typically means the use of open networking where the customer can choose the hardware and software separately. Commercial ASICs can now deliver the performance and functionality needed to build a switch for carrier environments. White box switches running Network Operating Systems (NOS) are running all of the control plane software on the CPU of the white box switch. The CPU and memory are limited and, as a result, so is NOS performance. The deeper issue is that these are still managed as appliances. As the number of routers increases, this will wreak havoc on OpEx – there simply aren’t enough trained technical staff to manage that kind of growth in appliances. Thus, we are seeing the limits of the appliance model even in disaggregation.

Volta takes a different approach to router disaggregation. In our platform, we run the compute intensive control plane software in the cloud which allows us to run up to 255 virtual routers (VR) on a single switch. Each VR is its own administrative domain, so they run independently. This will be critical for RAN sharing and network slicing. We also change the OpEx equation because we use APIs and service models to facilitate carrier grade automation. We believe the future is disaggregating the appliance model.