As we have described, network operators and cloud providers are looking at using NFV software and industry-standard hardware to deploy networking applications as VNFs rather than using integrated, proprietary hardware and software systems. This means that just about any application can be deployed with NFV software on top of a standard hardware platform.
This makes the possibilities somewhat endless, and the market for network functions –most of which can conceivably be migrated to software – quite large. It is certainly billions of dollars, if you take a look at traditional network applications such as firewalls, load balancing VPNs, application delivery, mobile evolved packet core (EPC), WAN connectivity, many types of security applications, and IP routing – all of which could be moved (and are being moved) into a VNF model.
VNFs vs. Purpose-built Hardware
The standard question that a network operator might ask about VNFs vs. traditional hardware-based functions is “does it scale” and “is it carrier-class.” The proprietary hardware industry has been formed for many decades and many companies have spent years optimizing their gear, including using custom chipsets, to boost performance. In addition, existing OSS and BSS systems are geared towards physical systems, with the use of VNFs, automation and orchestration play a significantly bigger role with managing the entire lifecycle of VNFs, including creating new instances as and when performance demands it.
Organizations looking at VNFs will have to make careful evaluations about whether or not the VNF measures up to the performance and stability of the function they are targeting as it runs on a purpose-built box. In early POCs, we’ve heard from end-users that some VNFs are poorly-executed as simple ports from physical boxes to a virtualized instance with capacity, throughput and stability issues. As well, the many variations of architectures to extract maximal performance from the underlying NFV-I, ranging from PCI pass-through, SR-IOV (single root I/O virtualization on PCI buses), DPDK (Intel’s Data Plane Development Kit), specialized off-loading to NICs (network interface cards) and other acceleration techniques create a lot more complexity than most organizations are willing to sign up for. We expect these teething pains to eventually be overcome. In the meantime, organizations should also examine the feature-set and performance characteristics of VNFs and how they compare with their physical counterparts.