Today Ericsson is working with 195+ customers around the globe for NFV transformation, and we know that this is not as easy as we expected.
Also, the promise of NFV from 2012—lower CAPEX and OPEX, increased operational efficiency, no vendor lock-in, and shorter time to market—has not come true. Instead we are facing complexity, integration difficulties, lack of standards, operational challenges, and more.
However, some early adopters, who have been working very hard for the last 3–4 years, are now starting to come into live operation with 10–20% of their traffic running on their new virtual networks. We have about 60+ customers in some sort of live commercial operation.
We see generally two types of deployments. The first is a single node with a virtualized silo, for example, vEPC on COTS-type of hardware. These operators can even have several vertical systems, for example, virtual infrastructure managers (VIMs) from different vendors on different VIM types, but OpenStack dominates. The other main deployment is a common cloud or NFVI with a few different VNFs from different vendors. This type of configuration has proven more difficult to integrate and also requires an operational transformation with separate operational domains for NFVI and VNFs to benefit from the cloud.
5G will require a virtualized core on a telco cloud
With 5G, we are introducing concepts like network slicing and distributed cloud. New services for industries will benefit from this with, for example, applications for industry automation and manufacturing, which are likely to come much earlier than the more famous self-driving cars. These 5G features will require a virtualized core on a telco cloud. Considering the increased complexity, much automation will be required, which underpins the argument further.
As the complexity increases, there will be a great need for increased automation and orchestration and we also see more interest from the market. Orchestration means also service orchestration, where customers such as enterprises order entire services such as SD-WAN, IoT, or enterprise MBB online in a self-serve portal. The services will utilize network slices that can be deployed via TOSCA models in the service orchestrator.
What can we learn from the early deployments?
NFVI platform integration.
The challenges of integrating an NFVI stack including open source software with VNFs on top is a challenge, because the number of possible combinations is huge and it’s not possible to test everything beforehand. We are, therefore, seeing a trend towards choosing more pre-integrated configurations in partnership with a vendor. Even if this limits the combinations with the benefits of short deployment time and lower risk, it’s still an open solution based on open source and COTS hardware. The NFVI platform must, of course, be prepared for an evolution to cloud native applications which will increase as we introduce 5G.
Operational transformation.
This is a key area that is seen as the biggest hurdle, according to reports from several recent analysts such as IHS. Those who start to plan a transformation early and succeed will be in a good position going forward. A big change is to go from today’s vertically siloed operations of nodes or domains such as EPC, IMS to have a separate team that operates the NFVI, serving all applications. This will be key to reap the operational benefits and to be able to automate more-complex structures such as network slices. A good example here is Swisscom, which has already come far on this transformation and has services in operation. This will also enable a more efficient DevOps operation with small frequent updates, as opposed to today’s yearly upgrades.
Security.
This will be emphasized even more with NFV and 5G, and the experience here is that this must be planned for from day-one. If not, a lot of extra rework has to be done later on.
Do you want to hear more?
View a number of recent NFVI presentations on these topics in the 15–20 min short webinars.
