So you are in a sprint mode for deploying virtual CPE ( Customer Premises Equipment)?
You, as an operator, want it FAST. As you want to cut your CAPEX/OPEX or perhaps you want to provision services faster.
But, something is holding you:
Perhaps you are confused by the industry’s discussion on what is the right strategy for deployment. I.e. should you place virtual CPE functions on a customer site OR run them from a cloud in a central location?
After all, different vendors approach you explaining what is the best approach (in line with their product strategy)
But, let’s face it!
ETSI specifications allow placing virtual CPE functions anywhere. They can be located anywhere as long as the functions serve their purpose.
Now that all approaches are correct, what strategy you should follow?
This blog attempts to make a point that it is best for an operator to deploy virtual CPE in phases. They can start with the least disruptive approach and then move slowly to a full-scale NFV.
To understand this approach, let’s bring up the discussion of the location of virtual CPE, again.
Where to place virtual CPE?
Should virtual CPE reside at customer site (also called Edge vCPE) with all the virtual functions (VNFs) including routing security, firewall, WAN optimization placed there, perhaps running on an x86 server or a white box?
Or a Centralized virtual CPE strategy (also called Cloud vCPE) is better? When an operator’s data center/cloud hosts all customer functions.
Or a third approach which is also called Hybrid virtual CPE ? That is, some functions run from a customer site and moving all other functions to the operator’s cloud.
What is the right approach to virtual CPE deployment?
On the face of it, the Edge model seems attractive.
First, from a cost point of view, it is based on the pay-as-you-grow model. Invest at the customer site, once business comes. This is in contrast to Cloud model; which needs some scale of the data center (x86 servers) established from day one as a shared resource for an unforeseen number of customers.
Second, the operators are already used to hardware CPE at a customer site and virtualizing the same functions on x86 server at the same location may be easy to accept and least disruptive for the operation and engineering people. Not to forget that the performance would be predictable in terms of latency, QoS or security (which may be an issue if all the functions are moved to the cloud).
Third, unless there is a full fledge Management and Orchestration (MANO) in place (as the standards work is still in progress), there is a risk of going with the vendor preferred open interfaces for MANO. There is no guarantee that these open interfaces can gather industry momentum in future or become part of a larger ecosystem that you can benefit from. This risk is somewhat mitigated with an edge model as you may not need full-scale MANO initially. And the vendor’s experience with the FCAPS (Fault, Configuration, Accounting, Performance, and Security) for the physical CPE can help him manage the virtual CPE.
Come to think of it!
NFV is still in its infancy. The FCAPS standards are not yet finalized. Selecting the centralized or the Hybrid approach would mean running elastic services in the cloud and going to a vendor’s specific management and orchestration (MANO). In this blog here, I argue the risks of going as an early adopter for NFV.
If you think NFV is carrier grade today, think again!
Opposition to the Edge Model
Let’s see the second side of the picture.
The edge model is opposed by some. Some even question that Edge model is NFV in the first place? They argue that NFV needs to run elastic services in the cloud with full-scale MANO with no geographical limitations. They say that the direct benefit of putting functions in the cloud is to take benefit of shared resources of x86 servers in the cloud. Shared infrastructure means decreased CAPEX and OPEX. By running all services on the edge, an operator will lose such benefits.
Hybrid model may be the ultimate goal
The industry is reaching a consensus that a Hybrid model might be the best one, where some functions (that need performance, less latency, security) need to reside at a customer site and other functions can move to the cloud.
However as mentioned earlier going full scale Hybrid today means that the operator needs to be aware of the performance of the system in the cloud.
When you have a mix of elastic third party virtual VNFs that are chained in the cloud, you need to be sure about performance, delays, QoS, reliability and fault management. We are not talking about virtual machines in IT clouds but carrier grade telecom infrastructure. Even the open stack is not fully carrier grade today.
Why not start small?
Start at the edge and move to the Hybrid model
Go with a pay-as-grow and the least disruptive model.
Start virtualizing functions at the edge. This needs less testing, fewer interoperability tests, and less initial investment. This may not even need a full-scale management and orchestration. The customer will have its dedicated hardware for virtualization. As it is not a shared resource so it is much easier to implement, much easier to maintain, much simpler to troubleshoot.
Then slowly move the functions to the cloud. But for that to happen, you need to make sure that you have already selected a vendor who also supports Hybrid model.
While your customer enjoys the service at the edge. you can keep on testing the performance of the Hybrid NFV in the cloud.
Then you can move to a full scale Hybrid NFV when you are sure that:
– You can run the vendor’s own virtual functions and third party virtual functions without any performance and fault isolation issues.
– You know that you can introduce full-scale Management and Orchestration with a peace of mind that the open interfaces by the vendor have already developed a bigger ecosystem or they are already standardized by the standard bodies.
Remember that MANO is your most precious investment in NFV. virtual CPEs can come and go but you may be stuck with the MANO for a long time.
You can aim for Hybrid NFV but better to take it in phases. Start virtualizing at the edge and then move to a more Hybrid model. Dealing with clouds is not predictable.
Take your time to test, test and test. In the labs and in the production.
Take NFV as a marathon and not a sprint. You may lose running too fast.
Finally I may request to tell me your take on this issue !