As the 2014 pipelines, business plans, and requests for proposals begin, now is the time to take a look at the state of the software-defined networking (SDN) and network functions virtualization (NFV) markets as a whole.
While nobody can credibly claim massive, across-the-board, worldwide deployments of SDN/NFV just yet, we are also beyond theory and speculation and well into extensive production deployments in leading carriers. To properly examine where we are, we can take a look at adjacent subdomains of carrier SDN/NFV, such as virtual networking, virtual functions, and management and orchestration (MANO, as defined by ETSI). What is in place and what is still being supplemented and analyzed? Are we “walking,” “jogging,” or “running?” What’s ready for prime time and what still needs work? Below, we dive in to each of the aforementioned subdomains to discuss.
The proverbial tug of war has been won; most SDN network virtualization efforts have adopted distributed overlays as a sustainable, scalable, and resilient architecture. This is the case for most service-provider environments, except for those historically driven by centralized circuit controllers (Infiniband HPC clusters or Sonet-type optical circuits, for example). Network virtualization overlays (NVO) use any existing IP network to map and capture subscriber flows to network functions. This has been proven in large-scale deployments and is being standardized at the Internet Engineering Task Force (IETF) for encapsulation and mapping formats, as well as for service chaining and stateless segment routing underlays. We also see significant efforts in adding this technology framework to SDN open-source stacks like OpenDaylight, which is a positive sign for the industry.
Networking Virtualization Overlay (NVO) Supplements
For many implementations of NVOs, we still see many in-host (hypervisor) rather than the needed in-network, OpenFlow-based NVO implementations. This is true for most of the offerings except a few carrier-focused SDN players and SDN-adapted fabric players.
In the long run, host-based approaches alone are not sustainable. For carriers, network virtualization edges (NVEs) should aggregate hosts and virtual and legacy functional endpoints, and be kept in reasonable numbers, e.g. hundreds to thousands, not hundreds of thousands. NVO mappings today adopt Hadoop, DHT, and distributed directory practices for global lookup, but even these technologies have consistency/performance tradeoffs with respect to number of nodes. On the positive side, we are beyond the “spray and pray” multicast proposals for distributing global information; these would probably result in inconsistency quickly. That being said, mapping authority for resolving lookups in dynamic carrier applications will need to be further optimized, including potential hardware acceleration of lookups.
We have seen a lot of examples and use cases in this area, many evident in original proofs-of-concepts (POCs). Clearly, the domain leading the market is that of mobile applications, a.k.a the GI-LAN functions. This is because these functions are at the intersection of “painful” and “attainable.” These functions are relatively new and are dynamically evolving along with the proliferation of smart phones. Traditionally, they had been running on servers and do not have clear structure governing their behavior or methods for weaving into the data-path. In this category we include filters, analytics, optimization, tagging, transcoding, and other prolific functionality. There is no one player dominating all the different disciplines involved, therefore NFV/NVO is a good practice to structure and package these into a system.
A more formal set of network functions being virtualized quickly is the IP Multimedia Subsystems (IMS) and session border controllers (SBCs). Although these are subject to strict 3GPP standardization, the fact they are based on SIP (an open, XML-based protocol), makes it easier to repackage to virtual machines and globally map user flows to using SDN/NVO along with light application-level gateways (ALG).
We still have not seen enough progress when it comes to virtualizing access gateways. In the fixed access aggregation front or BRAS, protocols are simple enough, but there is not enough market momentum to trigger any actual efforts beyond conceptual trials. Evolution of carrier Wi-Fi offloading and the Internet of Things (IoT) may help us move beyond the experimental stages. In the cellular mobile access aggregation, or the evolved packet core (EPC), the complexity of the 3GPP stacks and the need to maintain global international roaming interoperability puts the ball in the court of those who have these 3GPP protocol assets in production today. These groups are not yet inclined to turn these assets over to yet another VM hop in the function chain of an Open Compute-based carrier environment. However, this transition is inevitable, and open-source 3GPP stacks like OpenIMS and OpenEPC help release the lockstep in the industry and move this along.
Management and Orchestration (MANO)
Here we too have a clear winner. OpenStack is getting increasingly “installable” and stable. OpenStack allows carrier organizations to create their own Amazon Web Services (AWS) types of environments for hosting virtual functions, and not surprisingly, they are doing just that. The accumulated experience of the OpenStack environment from the large enterprise market helps it mature faster and readies it for primetime, as was and is the case with the virtual-compute hypervisors and NICs. This has been a major advancement in carrier-specific evolution.
We still have quite a ways to go for prime time; however, significant efforts are being applied in multiple axis of the industry. First, the ability to use and integrate OpenStack with NFV requires specific carrier augmenting beyond that of generic VMs. OpenStack for NFV also needs to integrate with the underlying NVO virtual connectivity and efforts to integrate NVO mapping into OpenStack Neutron. To complete the triangle, an integration to existing carrier OSS/BSS Diameter and Gx interfaces is also required, both to OpenStack for NFVs infrastructure and with NVO Mapping authority or NVA for connectivity, chaining, access control, balancing tenancy, etc.
While the above is an abbreviated account of the current state of carrier SDN/NFV, we are in the midst of transitioning from a walk to a jog on this networking treadmill. As the year progresses and more of the above elements fall into place, I expect we’ll eventually up our pace to that of a run and exceed expectations for 2014…without me even having to lace up my sneakers.