Service providers are embracing NFV to drive profitability by lowering costs and creating innovative new services. In doing so there are many good reasons to push the resources for NFV-based Layer 3 service to edge of the network. There is also a need to also deliver Layer 2 Carrier Ethernet 2.0 (CE 2.0) services, both for the end user as well as for the infrastructure for Layer 3 services.
The question is: What is the best way to support both Layer 2 and Layer 3 services at the edge of the network? Following are some options.
Virtualization using discrete devices
One possibility is to use two devices: An open server and an Ethernet Access Device (EAD). In this case the open server would host any Layer 3 virtual network functions (VNFs) while the EAD would terminate the network traffic and provide for Layer 2 functions.
- This approach has the benefit of relying on existing devices and is consistent with the ETSI NFV ISG.
- The performance of existing EADs is well known and deterministic.
- The obvious drawback to this approach is two separate devices are required.
Virtualization using a hybrid device
Another approach is to combine the hardware functionality of an EAD with that of an open server. This could be implemented in several ways:
1. An existing EAD could be augmented with an add-on pluggable compute module.
2. An existing server could be augmented with a plug-in EAD contained in a SFP.
3. A hybrid EAD/server could be constructed with separate compute and EAD elements. This approach is similar to #1 above, but has the advantages of higher-performance connections between the EAD and compute elements, as well as using a larger and higher performance compute node.
- The hybrid approach is an improvement over separate components in terms of units, size and cost.
- The hybrid approach eases the task of ensuring deterministic packet delivery by use of hardware for forwarding.
- This approach is necessarily limited in one of its aspects. Either in its compute power and flexibility, or in its ability to change how its EAD function behaves.
The previous two approaches are evolutionary. We are taking the revolutionary stance of advocating that if you are going to virtualize, go all the way. In this case, we need to virtualize the EAD, so it can run with the other service VNFs.
As shown, the virtual EAD is running on a standard open server. Some advantages of this approach:
What about performance?
One potential drawback is performance. Can a software implementation of CE 2.0 functionality provide the needed performance in terms of throughput, loss and latency? The answer is yes -- if you use the latest tools and technologies, with performance in mind. We recommend that you:
Comparison of approaches
The table below compares and contrasts the approaches described above.
Table 1: Attributes of the Different Approaches
|Approach||Separate Devices||Hybrid||Pure Play|
|Consistency with ETSI NFV||Best||Good||Best|
|Layer 2 Throughput||Best||Best||Yes (optimized)|
|No (open vSwitch)|
|Scalability||Best||No (requires redesign)||Best|
|Applicability to Cloud||No||No||Best|
Go virtual, all the way
The vision for NFV is to enable the use of standard and open compute servers to host a variety of mix-and-match software VNFs. To date, the focus has been on Layer 3 services and VNFs. To fully achieve the NFV vision means delivering these VNFs and services in a scalable, cloud-based manner to all parts of the network. That means addressing the question of Layer 2 functionality.
As I have described above there are three different ways to provide Layer 2 functionality. While the discrete and hybrid approaches have some advantages (especially in the near term), fully achieving this vision requires inclusion of Layer 2 capabilities as a software module rather than as a hardware adjunct. Doing so enables complete flexibility in delivery of services at any part of the network, and ensures that operators will be able to ride the technology curve of open servers. The result is a simultaneous lowering of costs and enablement of new services, which combine to drive profitability.
— Prayson Pate, CTO, Overture Networks Inc.