Back in 2012, just after the NFV founder-operators had published their seminal white paper, I interviewed the CTO of a communications service provider (CSP) about his reasons for supporting network virtualization. He pointed out that it was perhaps 30 to 40 times more expensive to carry out a security process in a chassis-based firewall at the network edge than it was to process the same function in the core. "So there has to be real economic justification to distribute physical network appliances beyond generic packet processing," he explained.
In other words, his large enterprise customers needed to have deep pockets and a rock-solid reason to pay for them. As a result, about 50% of this CSP's customer base bought basic NAT-related and firewall capabilities and only around 15% invested in WAN optimization or load balancing.
NFV, the CTO suggested, would at last enable the operator to deploy a broader range of virtual network functions (VNFs) across its footprint, on a common compute platform, "at marginal cost." He expected significantly to boost sales of Layer 4-7 network services as a result, and called the business case for NFV a "no-brainer."
The opportunity this CTO described is now known as the "virtual CPE" (vCPE) use case. At first it was thought that VNFs associated with vCPE would run in the cloud at the provider edge, but vCPE deployments increasingly involve a mix of VNFs running both on the customer premises and at the edge of the network. Increasingly, too, operators don't just want to expand sales to existing large enterprise customers; they see vCPE as a way of cracking the elusive small and midsized business (SMB) market, under-served distributed branch offices and residential customers, for whom a range of Layer 4-7 services, while badly needed, have remained well out of financial reach.
Early vCPE deployments have been disappointing, however. It turns out, for a variety of reasons, that if a VNF vendor simply swaps a dedicated hardware environment for a virtual machine (VM) implementation of a network function, there are not many cost benefits after all. Thus, the business case envisaged by the CTO more than three years ago has proved something of an illusion, at least when it comes to the cost-sensitive SMB, distributed branch office and consumer markets.
In fact, virtualization has highlighted these segments' reluctance to pay for large, monolithic pieces of network function when they don't need most of the features. So the race is on to find VNFs they will pay for and on which CSPs can make a profit -- open source VNFs, for example, which also tend to package network functions more simply, a single feature at a time. With atomic VNFs, it's easier to give customers the choice to buy only the specific feature(s) they want.
There is a natural affinity between small, single-feature VNFs and next-generation approaches to cloud application development and virtualization. Scale-out, microservices-based, cloud-native applications and containerization were considered esoteric in 2012, even by the IT community. But these technologies are maturing fast and adoption rates are soaring.
When applied to the vCPE use case today, the results are particularly intriguing: It turns out that scaled-out, single-feature VNF instances can be deployed inexpensively in containers, where they can be chained at application level with multiple other VNF instances. Each container can hold a different collection of VNF instances, tailored to the needs of a particular customer. CSPs can therefore at last serve their vast SMB and consumer bases with the network features they want, at a price point they can afford.
A key capability of a container-based vCPE platform is automated management that supports the rapid onboarding of new VNFs, the assembly of single-feature VNF instances in customer-specific containers, and the lifecycle management of those instances at huge scale. This requires clever tooling, as described in a new white paper, "Making a Compelling Business Case for Virtualized CPE."
And of course, the vCPE platform will need to offer the carrier-grade performance and high availability that CSP customers expect. This approach does seem to offer the breakthrough that early NFV supporters envisaged over three years ago, enabling CSPs to reap the benefits of enhanced service agility, reduced capex and opex and the provision of a differentiated service experience.
— Caroline Chappell, Practice Leader, Cloud & NFV, Heavy Reading
This blog is sponsored by ClearPath.