French philosopher Jean Baudrillard, best known for his seminal work “Simulacra and Simulation”, was fascinated with the interplay of “real” and “virtual”. He outlined a four-stage progression of simulation (or virtual): from the mere imitation of reality, over stages where virtual distorts and/or hides and fights reality – to the final stage at which virtual and real become indistinguishable.
This is an interesting progression to consider when we look at network functions virtualization (NFV). It opens questions: are virtualized network functions (VNFs) of today (still) a mere copy of their “real” counterparts (physical network functions or PNFs), or have they evolved to become a new class of software-defined products? Have we arrived at the point where performance, reliability and features of the virtualized products are indistinguishable from the physical ones?
The telco cloud market has seen great progress from the initial idea (which came and was mainly driven by operators), early discussions, definitions of frameworks and use cases, over R&D-focused activities, proof-of-concepts and multi-vendor NFV interoperability tests – to important field trials and many NFV deployments in production networks.
This progress worked faster for some network functions than others: for example, for network functions related to packet core, IMS and VoLTE/VoWiFi, and for those from service and application domains (such as CSCF, TAS, ENUM, HLR, HSS, MSS, MRF, PCRF, SBC, to name a few), the demand has almost exclusively been for virtualized implementations.
IP Virtualization at 2017
IP routing virtualizationmade good progress from the initial “low hanging fruit” applications – such as virtualization of control plane functions (e.g., route reflector), to advanced software-based implementation of advanced IP routing functions such as Broadband Network Gateway (BNG), Network Address Translation (NAT) and Provider Edge (PE).
Largely, this evolution in IP has been a result of important underpinning technological advances in processors, server technology and acceleration techniques which combined enabled much improved data plane capabilities, but more importantly a result of cloud-native thinking and new approaches in software design and implementation. With the cloud-native approach and cloud-focused software design (as opposed to just mimicking the PNF development), virtualized IP products became cloud-optimized and stopped being just virtual(ized) “imitations” of PNFs.
Virtualization of IP functions is happening in many areas in the network. Firstly, there are route reflector and “classical” PE applications, allowing linear scaling of IP routing functionality in a cloud environment. The notion of linear (cloud) scaling has a two-fold appeal to operators: it allows more cost-optimized entry points, as operators can start with systems that cater to immediate needs and do not necessitate deployment of bigger and more powerful (and more expensive) physical routers. Also, linear scaling capabilities mean fewer hassles with vertical scaling, allowing elasticity to be achieved by addition (or deletion) of resources on need-to-grow basis and with no service interruption.
Virtualized PE is finding its place along PNFs or in new, virtualized-only overlays—in fixed and mobile environments, enabling new and improved services for residential markets and dynamic enterprise services. A growing number of data center architectures are deploying virtualized routers for DC gateway applications.
Then, there are virtualized IP implementations of diverse IP gateway functions. Broadly speaking, these are IP router applications where some sort of translation is required between a large number of user or device sessions (IP flows) coming from the access side, and high-bandwidth broadband connectivity towards the network or to the internet. Example here is a virtualized Broadband Network Gateway, which is quite appealing to operators interested in distributed BNG applications, which is often combined with L2TP Network Server (LNS) functionality – providing next-generation broadband access aggregation platform and allowing agile provisioning and ability to flexibly adapt to arbitrary mix of retail and wholesale scenarios.
A virtualized security gateway (SeGW) is another example, used for management and termination of large number of IPsec sessions in fixed, mobile and converged networks.
Along with virtualization of IP-based mobile packet core functions, we have seen the introduction of virtualized WLAN gatewaysfor full operational integration of carrier Wi-Fi with cellular, ensuring seamless wireless connectivity with much improved user experience. The virtualized implementation of a WLAN gateway allows much faster and more flexible deployment with other virtualized packet core functions (e.g., mobile gateways, mobility management).
In addition, there is a growing number of virtualized implementations of protocol translation such as Network Address Translation (NAT) and Mapping of Address and Port using Translation (MAP-T) Border Relay functions, addressing the important evolution and coexistence of IPv4 and IP6 network segments and domains.
Lastly, we have witnessed the emergence of virtualized IP implementations for value-added services and specialized applications such as application assurance (including DPI), analytics or telemetry.
Lessons Learned, Challenges and Opportunities Ahead
A typical journey into IP virtualization comes in the form of careful steps from vendor demos, lab trials and interop tests, over limited commercial trials, to full commercial deployment. Most of initial challenges revolve about the deterministic performance: in a large NFV ecosystem with a lot of “moving parts”, VNFs can become dependent on many factors such as server types, available CPU cores and system memory, and on whether and how many other VNFs are concurrently running. Commonly, operators would pick a single VNF and “play” with it, until they learn more about it and discover all the important issues that they need to look out for. The next challenge here could be a predictable, deterministic performance of several concatenated VNFs – a service chain. Once a VNF or a group of VNFs has been fully benchmarked and infrastructure requirements tweaked to assure predictable performance levels, other considerations need to be made – like management and orchestration (MANO) design, and further integration with OSS/BSS.
At this stage, the focus would shift to areas such as flexibility, programmability and automation.
Management systems must ensure flexibility and operational continuity across physical and virtualized network domains. In addition, cloud management platforms must create a solid foundation for deployment of other virtualized applications.
A hybrid network has been one of big “discoveries” of virtualization – the fact that not all the functions can and will be virtualized and that there will be a need for high-performance physical platforms for quite some time (let’s just look at core routers). This understanding drives a need for a balanced approach and a right mix of PNFs and VNFs in the network; the question is not about a choice between a VNF and a PNF, but about having the right platform (physical or virtualized) with right capabilities at the right place in the network (read: the right network function) at the right time (in operator’s evolution).
The end-goal of NFV—the one which promises most benefits—is automation, which takes many forms: from the “natural” and obvious SDN programmability to automation based on and employing network intelligence and advanced self-regulating frameworks which improve network resiliency, agility, and overall efficiency. Using network’s “big data” for insight-driven automation becomes a “Holy Grail” of virtualization and the area where most of the challenges and opportunities will lie ahead.
5G and Virtualization
5G has become the hottest topic in the industry with a wide range of issues: from spectrum and related regulatory and licensing issues, over new radios (5G NR) and new packet core (5GC), and options for integration with existing LTE networks, to network slicing—partitioning of mobile network resources for a specific application, use case or customer, and further evolution from all-IP to service-oriented architecture (SOA).
5G also brings the focus on new and improved use and business cases (such as 5G-enabled enterprises, industries and vertical sectors, smart cities, fixed-wireless access for rural broadband) – which are needed to justify huge investment in 5G and make it a success.
Many already agree that 5G is a trigger for much larger network transformation, not just affecting radio access network and packet core, but driving the complete overhaul of the network and requiring significant changes across mobile transport, cloud and application and service fulfilment domains.
With 5G, there will be new areas of virtualization: from the RAN (e.g., virtualized baseband processing and creation of radio clouds), to further virtualization of the packet core and application domain. The growing interest in video, AR/VR and low latency applications is driving Multi-Access Edge Computing (MEC) – with its distributed packet core, and content in mobile cloud centres which are closer to users (and devices). Virtualized MEC deployments will complement virtualized RAN and core, enabling a wide range of innovative and ultra-broadband use cases – for example, creation of temporary or pop-up networks or disaster recovery networks (in case of 5G for public safety).
While the physical nature of mobile transport (fronthaulandbackhaul) imposes limitations on what can be virtualized, there are many IP functions related to transport that will benefit from virtualization in mobile – from generic IP functions like PE and NAT to specific ones – like security and WLAN gateways. This will create many new opportunities for virtualized IP solutions that will reside in the same cloud centres which will be hosting other virtualized mobile network functions.
Real *and* Virtualized
Going back to Baudrillard’s four stages of interplay between virtual and real from the beginning of this article, it seems that we indeed are at the stage where it is indistinguishable whether some IP routing functions are implemented on purpose-built platforms or delivered as virtualized and running as VNFs on standardized compute servers.
However, there is more work to be done – in virtualization of more of vast set of IP functions, with challenges equally laying in the proper cloud-tailored software design, as in management, orchestration and automation.
At the same time, the realization that there still is (and will be for a foreseeable future) a whole range of high-performance, large-scale IP functions which require custom, high-performance and optimized hardware platforms brings into perspective the questions of choice and balance between the best and most cost-optimized platform for a specific network role. For that reason, network operators need to accept the hybrid world ahead in which physical and virtualized platforms will coexist for quite some time, and complement—not compete with—each other.