On July 18th, AT&T, a mobile network operator in the USA, rolled out Network Function on Demand, a Virtual CPE (vCPE) service, followed by its rival Verizon who also launched the similar service named Virtual Network Service, three days later.  

 

So far, many debates have been held about "how to apply SDN/NFV technologies to operators' commercial networks", and dozens of use cases have been suggested as a result. Of all the use cases, Enterprise vCPE has long been recoginized as the most efficient case in terms of capex and opex. And AT&T and Verizon, the top 2 operators in the USA, commercialized SDN/NFV-based services.

 

What's Virtual CPE (vCPE)?

 

Today, enterprises have a range of equipment running, for example, a router from Cisco, a WAN accelerator from Riverbed, a firewall from Juniper and so on - each designed to serve its own purpose only. So, having them, which would mean installing, connecting to the network, setting up, operating, updating, and replacing them, apparently leads to high capex and opex.

 

vCPE provides ability to virtualize the functions of each network appliance on to X86 server. The server then mixes and matches software-based functions (VNF: Vitualized Network Function) as needed, turning existing Physical Network Functions (PNF) into VNFs like vRouter, vWAN Accelerator, vFW, etc.

 

An NFV orchestrator in operators' network can create and manage VNFs. So, for example, when an enterprise customer requests for firewall service, an NFV orchestrator in the network can simply add vFW software (VNF) on to the X86 server, eliminating the need for sending someone to install a physicall firewall.

 

 

Where to place vCPE?

 

One of the most frequently raised issues while implementing vCPE is its location. It can be:

• on-premise across an enterprise (Distributed vCPE) or

• at edge PoPs or COs within an operator's network (Centralized vCPE)

Both approaches have pros and cons in terms of first cost, per-customer cost, scalability for increasing customers or increasing services requested by each customer, application performance, local communication survivability during WAN access link failure, and so on, as to be discussed below:

 

 

Distributed vCPE

 

Using distributed vCPE, operators do not have to re-build a full-scale data center or redesign the network architecture as it allows appliances to still remain within the enterprise customer's premise (on-premise deployment). This translates into fast market penetration with a minimum early stage investment.

 

On the other hand, because this approach offers 'enterprise-specific' vCPE deployment (i.e. each enterprise has their own vCPE deployment), sharing computing power across enterprises is not possible. Due to this inability, each enterprise should make sure their compute nodes (X86 server) have capacity margin that is 'sufficient' to cover any service addition that may be requested in the future (i.e. addition of VNF), on their own. And that results in higher per-customer costs, one of the biggest weak points of this approach.  

 

Centralized vCPE

 

The greatest weakness of centralized vCPE is that it requires full NFVI (Switching Fabric, Server and Storage) in the operator's networks (e.g. edge PoP or COs) and thus needs major investment at an early stage.

 

Also, long distance between enterprise customers and edge PoP/CO in an operator's network may lead to poor performance of some latency-sensitive services, like WAN acceleration.   

 

However, as it allows for efficient NFVI resources sharing among many enterprise customers within the same network, per-customer costs can be drastically reduced.  

 

 

Then what would be an ideal option? Coexistence!

 

It is predicted a hybrid approach where the two can coexist will prevail, as adopted and proposed by AT&T and Verizon. The two operators have already commercialized distributed vCPE. Centralized vCPE is expected to be launched first by Verizon by the end of the year and then by AT&T next year.  

 

 

 

KT Case

 

In South Korea, KT, the #1 wireline network operator in the nation, has been exerting efforts to develop centralized vCPE models for the past year, and introduced the concept of this approach in conferences locally held in February and June.

The concept is still in the PoC process, and, of course, not ready for commercialization yet.

 

What's noteworthy about KT's proposed architecture is vCPE (NVFs) at COs is connected to each enterprise customer via SDN network, and this network uses white boxes and open source. That is, the architecture lets not only SDN switching fabric located at COs, but also the access network from the CO to CPE (physical CPE) within the enterprise be deployed as white boxes.  

 

The architecture, of course still in the PoC stage, is forecasted to have a huge impact on the conventional network architectures, and become a catalyst for introduction of SDN/NFV in the Korean operator's legacy networks.