You're probably familiar with network functions virtualization (NFV) and its virtual network functions (VNFs) as a means to provide software replacement of dedicated hardware, aka physical network functions (PNFs). The ETSI NFV ISG has formally described the NFV architecture and its use cases, highlighting the agility and flexibility of software.
What impact VNFs will have is less clear. “Software eats the network” is the term scaring every hardware design engineer. The need for their skills seems to be disappearing as coders develop software for each and any network function, which can then be easily instantiated on a standard server.
It all sounds very attractive. However, the network transformation towards software-centric networking is taking longer than expected – for good reasons:
- New network architectures have to be designed and optimized for server-centric software-based networks.
- Operational processes have to be redesigned as well as aligned with agile methods.
- Staff have to extend their comfort zone and embrace completely new technology domains.
There are also concerns that software running on standard servers will not provide the required performance at similar price points as dedicated hardware appliances. A commercial off-the-shelf (COTS) server is optimized for software applications, and its cost scales with performance. Can a VNF hosted on a COTS server be as efficient as an optimized physical network function? Can it meet the service requirements?
The essential question is the required service performance. If there is only moderate demand for processing power, a server can easily do the job. Economies of scale in combination with flexibility and agility support a strong business case for the software VNF approach.
But if there is a need for high performance such as maximum throughput, low latency, or complex processing, then the situation is quite different. The latest multi-core, high-performance processor technology provides the necessary server performance for demanding VNFs. However, it also takes care of basic but resource-intensive network functions such as forwarding, service assurance or security functions. This is obviously not the most cost-efficient way to consume compute resources.
The result is that there is hope for experienced hardware design engineers. Their skills and competencies will help solve the problem. Any software function which eats up a lot of compute resources and which is required 24/7 is a possible candidate for being moved into a supporting hardware function. Those functions include service assurance, probes, encryption, timing and gateway functions among many others. I'm predicting a short lifetime for the software versions of those functions before they become hard-coded again – driven by economics and a need to have compute resources for agility rather than for static network functions. I refer to this transformation as reincarnated network functions.
The data center architects have already identified this option for performance optimization. A lot of functions are moving from software running on the server into hardware residing in network interface cards (NICs), boosting the performance of their server farms. Network functions such as flow management, tunneling mechanism, encryption or packet forwarding become implemented in FPGAs for higher speed and lower consumption of server resources. You might want to have a look at Microsoft Azure’s work on SmartNICs.
As their network colleagues redesign their networks using cloud techniques, they can build on the experience of the data center. Servers at the edge of the communication network will need to be optimized for highest performance at lowest cost. General network functions will move back from software into hardware, offloading the server and freeing up resources for revenue-generating services. Those reincarnated network functions will have a strong positive impact on the economics of the vCPE use case, especially with demanding high-value applications.
This is good news for our experienced hardware engineers. Their abilities will certainly be needed as our FSP 150 ProVMe edge compute platform complements virtual with physical network functions for best performance and highest flexibility.
Our latest extension of the portfolio even goes one step further. A pluggable server empowers our customers to seamlessly extend service offerings from connectivity services towards NFV-based managed services in the most economic and seamless way.