In-band network telemetry (INT) enables real-time and fine-grained network monitoring but incurs high transmission overhead. To mitigate this, encoding-based INT methods have been introduced to collect telemetry items using fewer bits than their original bits. However, prior works struggle to reduce encoding bit length when the magnitudes of telemetry items vary widely, as they rely on transmitting raw values. To address this challenge, we propose a prediction-based INT framework that effectively minimizes encoding bit length by collecting prediction errors instead of raw values. Our framework leverages both temporal patterns and inter-item correlations to robustly reduce prediction errors, thereby significantly lowering encoding bits while ensuring accurate reconstruction of original values.

Given the importance of robustness and resilience in emerging cloud-native networks, effective monitoring for fault detection is paramount and In-band network telemetry (INT) is a key candidate that enables real-time and fine-grained network monitoring with a programmable data plane. However, INT increases bandwidth overhead because network information is inserted directly into the packet header. In this paper, we propose a quantized INT (QINT) to effectively reduce overhead by considering the distribution of raw data. In QINT, the programmable switch encodes a raw telemetry item into a quantized bit stream using the Huffman coding scheme. To do this, QINT monitors the distribution of network telemetry items and encodes high-frequency data that are generated most of the time in a few bits. We implemented QINT on a programmable switch and our experimental results demonstrate that QINT can reduce the relative bandwidth usage by up to 60.6% compared to traditional INT, respectively.

In-network (or on-path) inference over programmable data planes (PDPs) allows fast and low-overhead inference using deep neural networks (DNN). To alleviate the massive processing and deployment cost of in-network inference, distributed deployment on multiple programmable network devices is mainly adopted. However, it is likely to produce a considerable amount of network traffic due to the exclusive forwarding chain and intermediate data between submodels. In this work, we propose a traffic-aware adaptive in-network inference scheme, TINIEE, to maximally reduce the network traffic of in-network inference without causing a significant reduction in classification performance. To this end, we first devise an adaptive inference method on the data plane striking the balance between the classification performance and the network traffic cost. Furthermore, we formulate a traffic minimization problem to decide the proper location of each submodel considering each flow's exit tendency with a predefined confidence threshold. Since the problem is excessively complicated, we devise a low-complexity practical submodel placement algorithm. We implement the proposed scheme on software-programmable switches, and the evaluation results demonstrate that TINIEE reduces network traffic by up to 34.48 % compared to the state-of-the-art, while maintaining sufficiently high classification performance.

In heterogeneous networks (HetNets), the dense deployment of base stations (BSs) often leads to severe signal interference. This interference causes radio link failures (RLFs) and ping-pong handovers (PPs), undermining the connectivity of user equipments (UEs). To address these problems, mobility robustness optimization (MRO) can be an effective solution. However, MRO operation without considering load distribution can overload specific BSs. This can increase signal interference and lower channel quality in high-load areas, degrading MRO performance. To mitigate these issues, we propose a load-aware MRO framework using multi-objective reinforcement learning. In the proposed framework, agents use two separate objective functions to learn the individual impacts of adjusting handover control parameters on both preventing occurrences of RLFs/PPs and distributing load. Through multi-objective learning, our framework minimizes the occurrences of RLFs/PPs while flexibly distributing load across BSs. This prevents additional RLFs/PPs and channel quality degradation caused by load concentration. Simulation results show that the proposed algorithm reduces the occurrence rates of RLFs and PPs by up to 27% and 52%, respectively.