Network Modularization, i.e., the refactoring of the functionalities of 5G Network Functions (NFs), will be an important tool to meet future requirements in the 6G System (6GS). In particular, via network modularization, it is possible to streamline the NF compositions and the inter-Nfinteractions. However, to achieve the highest possible efficiency without loss on inter-operability, we need a clear understanding of the different Key Performance Indicators (KPIs) and modularization techniques which are required to achieve the necessary end-to-end performance. In this work, we investigate the KPIs, the modularization methods prioritized by the research community, and the industrial implications thereof. We also outline the or-chestration requirements for the successful integration of modules to 6GS and the one possible integration of quantum technology to solve the synchronization challenge.

In our digitized society, emerging applications require highly-performing and flexible networks that can adapt to satisfy varying connectivity needs. P4 as a domain-specific programming language for data plane pipelines introduces the required flexibility through easy-to-use programmability. However, the performance of P4-capable devices is still an open question that has not yet been completely addressed. Understanding whether a P4-enabled device can meet the performance requirements for a specific network function pipeline is key for planning as well as for the selection of the proper deployment scenarios in a network. To bridge this gap, we propose a simple analytical model that can predict the performance of network functions written in P4 for a given device. The programmable data plane of P4 devices is modeled as a forward queueing system with a variable service rate that depends on the complexity of the configured data path program. On top of the data plane model, the controller’s interaction is modeled as a feedback queueing system. In terms of the analysis, we first assume exponentially distributed service times in the data plane and control plane. In a second step, we extend the analysis to generally distributed service times using approximations. In order to cover a wide rang of possible behavior of the control plane, three types of distributions with different coefficients of variation are inspected: Erlang, exponential, and hyperexponential. We evaluate the accuracy of our model for different scenarios and show that the discrepancy between actual results and our analytical predictions does not exceed 8.7%. We also validate the model with a commercial P4 hardware switch.

As public and private cloud-native deployments of the 5G Core (5GC) networks are rolling-out on a wide scale, attention is shifting towards efficient state management. While stateful deployments were the default method in the previous generations of mobile networks, they lack the necessary flexibility that cloud-native orchestration demands. Yet, traditional approaches taken to enable stateless deployments require the operators to sacrifice on performance due to the frequent state transactions. To overcome this issue, in this paper we propose a Piggyback-based and a Proactive-Push approach which allow for procedure-aware stateless 5GC systems. Our evaluations highlight the advantages of the Piggyback approach for two synchronous control procedures, reducing their completion time by ~44% and ~70% compared to the baseline. For asynchronous procedures, the Proactive-Push approach outperforms the baseline with ~13% and ~22%. More importantly, these mechanisms do not pose additional overhead on CPU and bandwidth utilization.

The introduction of the Service-Based Architecture (SBA) for the 5G Core Networks has drastically changed the way these networks are designed and operated. Aiming for higher flexibility and agility, the adoption of SBA is the first step towards cloud-native deployments of 5G Core. However, the high degree of functional decomposition in SBA has implications in terms of increased inter-NF signaling traffic during the execution of control plane procedures, as well as an increased complexity in orchestrating a system with tight inter-NF dependencies. In this work, we introduce PP5GS as a stateless 5G Core architecture that implements a procedure-based functional decomposition of the 5G Core NFs. We develop Per-Procedure NFs for four different control plane procedures and perform extensive evaluations in a private cloud environment orchestrated with Kubernetes. The results show that PP5GS requires up to 34% and 55% less computing resources compared to the baseline stateful and stateless systems, respectively, while generating at least 40% less signaling traffic. Moreover, complex control plane procedures can complete up to 50% faster. Lastly, the results show that PP5GS is a more feasible architecture in leveraging edge-offloading of 5G Core NFs.

The recent trend to deploy programmable packet processors in cloud environaxsxsments enhances the packet processing capability without losing the flexibility to adapt the functions at runtime. In particular, distributed edge clouds can have a heterogeneous programmable processing substrate made up of different classes of devices: CPUs, NPUs, FPGAs, etc. However, managing the allocation of workloads in such a heterogeneous programmable processing substrate, in particular deciding where to instantiate a certain function, is a non-trivial task with many decisive functional and QoS-related factors. In this paper, we propose a mathematical model for optimizing the embedding of Service Function Chains implemented in P4, while considering the functional and QoS requirements associated with embedding requests, and the various types of processing devices that have different properties in terms of processing delay and supported features. To satisfy delay requirements, the problem formulation utilizes performance models to predict the forwarding latency associated with different candidate embedding options. Furthermore, a greedy solution is proposed to solve the problem in an efficient manner. Finally, a detailed numerical evaluation is conducted to evaluate the formulated model when different workload and infrastructure characteristics are varied and to evaluate the effectiveness of the proposed greedy solution.

Software-Defined Networking (SDN) lays the foundation for the operation of future networking applications. The separation of the control plane from the programmable data plane increases the flexibility in network operation. One of the most used languages for describing the packet behavior in the data plane is P4. It allows protocol and hardware independent programming. With the expanding deployment of P4 programmable devices, it is of utmost importance to understand their performance behavior and limitations in order to design a network and provide Quality of Service (QoS) guarantees. One of the most important performance metrics is the packet mean sojourn time in a P4 device. While previous works already modeled the sojourn time in P4 devices with controller feedback, those models were rather simplified and could not capture the system behavior for general cases, resulting in a potential highly inaccurate performance prediction. To bridge this gap, in this paper, we consider the system behavior of P4 devices for the general case, i.e., under general assumptions. To that end, we model the behavior with a queueing network with feedback. As it is impossible to provide closed-form solutions, we consider different approximations for the mean sojourn time. We validate our results against extensive realistic simulations, capturing different behaviors in the data and control planes. Results show that the most accurate approximation in almost all cases is the one in which the queues are decoupled and considered as independent despite the fact that there are dependencies. The level of discrepancy in the worst case does not exceed 18.2% for service times distributions with a coefficient of variation not greater than 1.

Languages for programming data planes like P4 sparked a plethora of new applications in the data plane. The dynamic, evolving environment makes it challenging to understand what performance can be expected when running a program in a specific data plane target. However, knowing this is crucial for network operators when upgrading their networks.

Networking devices with programmable data planes, such as P4 programmable devices, are gaining more popularity because of the flexibility they provide in describing the packet processing behavior. Despite this attained flexibility, the performance of these devices can be the Achilles' heel in case the desired performance level is not met. To this end, we evaluate the performance of three state-of-the-art P4 devices focusing on the following properties: (i) the device's processing latency as a function of a scaled number of flows; (ii) the device's response time in reaction to control plane commands. The scalability analysis shows that different devices have different limits on the maximum number of flows they can support. On the other hand, the device's response time to control plane commands is found to be in milliseconds, which is three orders of magnitude larger when compared to the measured data plane's packet processing latency.

Data plane programmability brings network flexibility to a new level. However, it introduces the complexity of the data path's program as a new factor that influences packet forwarding latency and thus devices' performance. Accurate identification of the relation between data path complexity and packet forwarding latency enables the design and management of networks with predictable performance. In this article, we leverage the characteristics of P4 programming language to provide a method for estimating the packet forwarding latency as a function of the data path program. We analyze the impact of different P4 constructs on packet processing latency for three state-of-the-art P4 devices: Netronome SmartNIC, NetFPGA-SUME, and T4P4S DPDK-based software switch. Besides comparing the performance of these three targets, we use the derived results to propose a method for estimating the average packet latency, at compilation time, of arbitrary P4-based network functions implemented using the surveyed P4 constructs. The proposed method is finally validated using a set of realistic network functions, which shows that our method estimates the average packet latency with sub-microsecond precision.

The advent of programmable network switch ASICs and recent developments on other programmable data planes (NPUs, FPGAs) drive the renewed interest in network data plane programmability. The P4 language has emerged as a strong candidate to describe a protocol independent datapath pipeline. With its supported architectures, the P4 language provides an excellent way to define the packet processing and forwarding behavior, while leaving other networking components such as the traffic management engine, to non-programmable fixed function elements, based on the capabilities of most programmable devices. However, network flexibility is essential to meet the Quality of Service (QoS) requirements of traffic flows. Thus, enabling programmable control for fixed-function elements like traffic management is crucial. Towards that end we propose the use of virtual queues in the P4 pipeline, investigate the application of virtual queue-based traffic management, and portability of the approach using different P4 programmable targets. Specifically, we focus on virtual queue based Active Queue Management (AQM) for congestion policing and meeting the latency targets of distinct network slices. The solution is compared to P4 built-in functionality for bandwidth management using meters, proving also that the additional dimensions of control are achieved without compromising the processing complexity of the solution.

P4 programmable data planes are becoming more popular due to the flexibility they provide in describing the packet processing pipeline. P4 successfully abstracts the processing pipeline of data planes using a limited set of constructs. The performance variation as a function of the configured P4 pipeline is an important aspect that should be studied. Analyzing the impact of different P4 constructs on packet latency helps in understanding the overall performance of P4 programmable devices. In this paper, we analyze the impact of a basic set of P4 constructs on packet processing latency to derive the influential parameters. We use the derived results to propose a method for estimating the packet latency of P4-based network functions implemented using the surveyed P4 constructs. Finally, we validate the accuracy of the proposed method by applying it to realistic network functions.

Abnormal driving refers mostly to drunk driving, fatigue driving, and aggressive driving behaviors. This work addresses drunk driving detection, which actually can be extended to apply to detection of the other abnormal driving behaviors. The approach learns about the related on-board vehicle sensors for detecting drunk driving behaviors based on learning using certain drunk driving cues. In a previous work, we developed a Hidden Markov Model (HMM) method and applied it to each time series of the selected sensors measurements. The prediction accuracy was highest for the longitudinal acceleration, with a maximum of 79%. Here, we extend our early work that was based on HMMs and employ Recurrent Neural Networks that specialize in time series, where our testing results indicate accuracy percentages that go into the upper nineties.

In this paper, we introduce a scheme of caching requested data in LTE networks. After data (eg., web pages) is cached after its first request, the subsequent requests of the same data are served from the network's caching memory directly. The idea in this scheme is new in that the caching memory is at the level of the mobile devices themselves and not in the LTE core network, as in other proposed schemes. Our approach aims to minimize the required changes in the functions of the various elements of the LTE network core for supporting the caching operations. As we illustrate in the results section, the proposed system results in lower overall data access delays, which are attributed to retrieving some of the requested data from Internet servers when such data are not found in the network.