The last thirty years have seen great strides in the deployment of communication network infrastructure. That growth is primarily due to the establishment of the IP protocol as a stable specification in the late 1970s. There have been many benefits to using an IP-based approach, and today all manner of different communications networks rely on its widespread adoption. That said, with an increasing emphasis on power-efficient communication for a growing variety of applications, there is a rising awareness that an alternative to IP might need investigation.
The Rise of TCP/IP
Since its formation, the TCP/IP protocol suite is responsible for creating the heart of the internet. The transport control protocol (TCP) deals with the transportation of data, the process of dividing up messages into packets and then reassembling them after transmission. The internet protocol (IP) takes care of the addressing and routing of the packets transmitted across a communications network of attached devices.
The TCP/IP protocol suite has seen many iterations to its specification over the years. Still, one of the guiding principles from its conception was that the endpoints occupied fixed positions and that their connections would remain unchanged. Increasingly, the networks in place today see that endpoints do move, and connections may be dropped and re-established depending on several criteria.
There are many different types of access technologies to create network connection sessions and an equally diverse number of handover processes in place. Market trends such as the industrial internet of things (IIoT) that are driven by state initiatives such as Industry 4.0, for example, involve the deployment of hundreds of resource-limited sensors, all requiring a network connection. The scale of the number of connections needed is staggering and is beginning to have significant implications for the connection approach.
Exactly how big the broader internet of things (IoT) will become is yet to be determined. So far, estimates by industry pundits and analysts as to the number of ”things” in the future vary in size considerably. That said, there is broad agreement that the number of nodes deployed is likely to run into the tens of billions. IDC, an industry analyst, has published a research report that forecasts an estimate of 42 billion nodes in operation by 2025. Further, the research predicts that the nodes will be creating approximately 80 zettabytes of data.
Streaming of video content is also rising at an alarming rate, driving data volumes even higher. Technical advancements that have made possible lightweight, low power, wireless-connected video headsets are driving the adoption rate of augmented and virtual reality (AR/VR) applications. Not only is this the case for gaming, but geographically-dispersed engineering functions are establishing virtual engineering teams that benefit significantly from an engaging, immersive video experience. As any gamer will tell you, latency is a big problem for gaming AR/VR headsets, impacting the user experience. Latency is likely to limit the wide market adoption of these headsets until a means of reducing it can be incorporated.
From the initial concept of TCP/IP, the number of use cases and the network requirements have expanded considerably. Today, each use case brings a set of distinctive requirements, including mobility, latency, capacity, bandwidth, determinism, power consumption, and quality of service (QoS). Use cases are diverse, from smart agriculture, smart health, IIoT, and the rapidly evolving autonomous vehicles.
It is no wonder that the provenance of using TCP/IP as the principle networking protocol moving forward might stimulate debate. TCP/IP belongs to an era when there wasn’t mobility, the available network bandwidth was limited, and security wasn’t the significant issue that it is today. The TCP/IP suite design criteria cannot accommodate many of the network requirements we now face, no matter how many tweaks and fixes we might consider.
Understanding the Shortcomings of TCP/IP
In addition to the shortfalls listed in the above applications and use cases, the motivation towards considering a non-IP-based networking protocol gathered speed as 5G started to become a reality. From the beginning of 4G implementation, the basic premise was that it always would be an end-to-end IP strategy, including the access and the core network.
But the approach taken with 4G is no longer appropriate for the 5G era. As mentioned above, the nature of network hardware is more diverse in terms of power budget, computing resources, and the all-important bill of materials cost. Adding TCP/IP stacks to any embedded device and provisioning the resource overhead required to manage them can be daunting for the developer and demand resource capabilities beyond what is required for the core application. For cellular IoT device nodes, this is particularly the case, requiring a lot of additional functionality, further increasing node complexity, and potentially impacting overall system economic viability.
For other applications, the concerns raised regarding the continued use of TCP/IP center on latency and mobility. In different situations, such as industrial robots and autonomous vehicles, the lack of a deterministic and ultra-reliable data connection is of significant concern. As in many use cases, the flaws with TCP/IP are only coming to the surface as other market drivers are advancing, bringing with them distinct requirements. Industrial robots can operate at a very high speed and achieve mm-level accuracy reliably and consistently, which dictates a predictable and deterministic behavior. The same operating conditions exist in autonomous vehicles where the lack of systems coherence and deterministic behavior can prove fatal.
Another significant change since the development of TCP/IP is the aspect of network security. The protocol suite’s vulnerability to denial of service (DoS) attacks is well known and, despite enhancements to strengthen its security capabilities, any network link is always considered an attack surface for an adversary. The necessary security enhancements to TCP/IP further added to its protocol overhead.
Interest in a viable non-IP networking protocol is now gathering momentum. In the main, this is to accommodate the many and various data services that are finding application, and downstream, maintaining network capital investment (CAPEX) and operating expenses (OPEX), these being vital financial benchmarks for network operators. In short, what is needed is a more rationalized approach that also addresses the technical concerns of latency and network security, as mentioned above.
There are network protocol alternatives to TCP/IP suitable for deployment. An alternative protocol can enable better traffic management, introduce the possibility of automated network configuration and optimization, and offer improved performance. Most of all, it shouldn’t rely, as TCP/IP does, on the user plane. TCP/IP has not had any alternative but to use the user plane. Instead, an alternative will put more emphasis on the use of the control plane for information such as packet destinations, dispensing with the need for addresses in packet headers.
For some use cases, functions such as address translation and header compression will no longer be required. Network mobility should become easier too. Coming back to our IoT edge node example, less protocol overhead processing reduces the processing load, reducing the node’s power consumption profile and prolonging battery life. Since the IoT appears to present many attack surfaces, using an alternative protocol to TCP/IP will also aid the defense against ever-present security threats from adversaries.
Non-IP Networking Protocols Explained
Each non-IP-networking protocol alternative has relative merits and inherent limitations. Applicable for 5G and already seen in LTE networks is the non-access stratum (NAS) protocol. NAS is suitable for non-IP transportation, mainly when the amounts of data are relatively small. Another advantage of NAS is that the data packets can take advantage of the security encryption features already in place.
Flexilink is another protocol alternative to TCP/IP. Currently, Flexilink is under development, but it embeds a label within the packet header that functions as an index into the routing table, making packet forwarding simpler to achieve.
The Flexilink protocol came out of an audio-over-ATM networking project at the BBC. Flexilink was initially best suited for live audio and video networking, but its operation addresses many of the limitations identified with TCP/IP. Flexilink works on the premise that a stream of packets is a flow. In a Flexilink network, the routing information is sent separately from the packets. Taking out the addressing information significantly reduces the overall size of the packet in addition to making it easier for a switch to forward packets.
Figure 1 illustrates the length difference between a UDP/IP packet and a Flexilink packet. Different addressing methods are also possible without the need to change the packet format. Most of the rest of the information about the packet also uses control plane messages; one sent for each flow rather than one for each packet. This approach immediately reduces bandwidth requirements.
Flexilink has two services (see Figure 2), each with a different packet format: AV & IT. Multiplexing techniques combine the AV and IT packets on point-to-point links. An AV service serves continuous media applications such as audio and video and offers the lowest levels of latency. Point-to-point links use an allocation period technique that sets a fixed latency and achieves a deterministic connection. The opposite is true for the IT service that suits more unpredictable and bursty traffic associated with computing processes and website browsing.
Some academic institutions are also working on non-IP networking protocol projects based around the recursive inter-network architecture (RINA). RINA requires far less routing, being more streamlined than TCP/IP, and also has excellent QoS, security and mobility attributes.
Setting a Framework for Standardization
The appearance of non-IP networking protocols needs careful handling. From an industry perspective, standards must be open and any thought of developing a proprietary solution avoided. Interoperability is an essential element too of any new protocol so that it can make use of legacy network infrastructure. The definition and promotion of a well-defined set of non-IP networking standards avoids any uncertainty throughout the industry.
During 2019, ETSI formed an industry specification group (ISG) to focus on the ongoing non-IP networking protocol research and standards development. With the goal to establish a viable and effective alternative to the long-standing TCP/IP protocol suite, the ETSI NIN ISG aims to have a testbed in place within the next 12-14 months. Once in place, the ISG will continue its focus to come up with a draft specification for the proposed protocol. The specification will address the latency, mobility, security and data volume issues discussed in this article in addition to maintaining compatibility. The standards definition work will include looking at making a more adaptable addressing model that can equally apply to both IP and non-IP nodes. Work so far with Flexilink network prototypes has established that it can interoperate with existing legacy TCP/IP infrastructure.
The standards definition that the ETSI non-IP networking ISG is currently working on will most likely initially apply to private mobile networks used for industrial factory automation and process control tasks. In the future, it is envisaged that the protocol standard will be applied to a broader range of applications and use cases.
Looking to the Future for Non-IP Networking
The growing interest in non-IP based networking points to the fact that TCP/IP will not continue to be the default protocol as it has been in the past. For the applications of today, with resource-constrained nodes and low bandwidth connections, TCP/IP presents a significant overhead, particularly when the amounts of data to be transported are relatively small.
TCP/IP was developed in an era when network endpoints were constant over time, and transported data was less time-sensitive. But the networking and communications industries are ready to embrace a new approach. Today’s technology involves high mobility, ubiquitous wireless connectivity, and the need for end-to-end security. Battery-powered nodes are now the norm and impose significant but necessary design constraints.
Non-IP networking promises to herald a new approach to packet routing that will quickly find adoption across different industry sectors and applications. With the capability to carry legacy applications’ TCP/IP traffic, the disruption will be marginal.
