Guest blog by Andreas Roeder, Solutions Architect, Service Providers
Even if there are not publicly available and massive across the board worldwide deployments of SDN/NFV yet, there is a lot of traction on the topic of stack consolidation in the industry right now. There are also a lot of adjacent subdomains for SDN/NFV such as the virtual networking itself, and how to connect all those pieces from a networking perspective - the virtual functions by themselves and the management and orchestration, which was recently defined as MANO by ETSI.
Let’s focus on the Virtual Function by itself within this blog. There are a lot of examples and use cases which have already resulted in some proof-of-concepts. It’s pretty clear today, that the main domain and target market are the mobile applications and the overall mobile operator ecosystem. Most of these functions are needed and they either belong to the ecosystem itself or even provide a premium differentiation role for the operator itself. In most of the cases they belonging to the ‘GI-LAN’ since the ecosystem is quite complex already, and will get more and more complex as soon as you need to integrate new functions to the already existing infrastructure.
There are 2 things which need to be accomplished here – it needs to be easy to integrate and provide a value inside the chain and it needs to scale. Some of these functions are not completely new and they have evolved over time. Traditionally, they have been running on dedicated on-premise equipment and didn’t need to follow a clear structure as they got integrated into the data-path of a subscriber. Typical functions of this category could be analytics, optimization, filters, header enrichment and manipulation, steering, tagging, and transcoding.
Its already clear form that this kind of stack could be spread among multiple VNF’s, maybe even a single VNF per NF with its own single stack, but this leads towards a very fundamental question – will it scale and will it result in any benefit? If the networks and chains will be built the same way like before, what will happen?
The number of devices/Hops/enforcement points will stay the same.
The overall complexity of the network will not decrease since the number of data-path elements will stay the same in the optimal case.
The number of network states is the same.
Each element which needs to keep state information and rely on a state machine will also be part of the chain afterwards which again will result in a higher complexity of infrastructure to manage.
The number of stacks will be the same
It’s not possible to reduce or standardize the number of network states in order to achieve simplicity if each NF will be performed in a dedicated VNF.
All of those points imply that VNF will be able to run with the same operational circumstance than that of the on-premise equipment today, which is likely, not the case. So the overall complexity might increase since the VNFs don’t have the same functional set and/or don’t scale today to the same limits like the box based solution today.
So having all this in mind, wouldn’t a better option be to consolidation the number of stacks and combine network state into a single VNF? Why not perform analytics, optimization, filters, header enrichment and manipulation, and steering at the same time in a single stack for a certain set of subscribers. This would lead to major simplification since it would reduce the elements in the path for a single subscriber dramatically.
Please reach out to the representatives from F5 at MWC - Booth # 5G11 (Hall 5) to learn more.