One of the recent methods for effectively utilising the available bandwidth of the network as conditions change is in-network packet trimming. It has advantages for applications which rely on low-latency and low-loss, such as video streaming. Packet trimming allows a network node to reduce the size of packets as they travel across the network, by removing some of the content, during the journey of each packet. In our previous work, we presented a system for layered SVC video streaming, and showed how the trimming process can be successfully implemented and utilised as a virtualized edge deployed function. If trimming is needed, there is a task which has to unpack all the video content and meta-data, including associated significance values, then determine which data to remove, and finally repack and rebuilding a new packet. Conversely, a simple scheme for packet trimming, in the form of a basic buffer cutting process, has been implemented in switches. This process keeps the first bytes of a packet, and drops the rest. Here we investigate if a buffer cutting process can be used for layered video, and describe the issues arising and the design aspects needed to adapt from packet rebuilding to packet cutting.

The rapid evolution of Large Language Models (LLMs) has revolutionized Artificial Intelligence (AI) applications across various domains. In particular, fine-tuning LLMs—essential for adapting models to domain-specific tasks—requires specialized strategies for distributed environments. In this paper, we present a novel framework for LLM fine-tuning in an edge-to-cloud continuum. By leveraging real-time network metrics and edge compute resources, an AI orchestrator dynamically manages workload distribution by selectively eliminating underperforming nodes to optimize fine-tuning efficiency. Experimental results demonstrate that our approach significantly reduces completion time while preserving model convergence, offering an energy-efficient solution for edge-enabled LLM fine-tuning.

The utilization of the packet trimming technique as part of 6G has previously been described. It allows a network node to reduce packet sizes, by removing some of the content, during the journey of each packet. This approach has been successful in many ways, but can be limited when there is very little bandwidth. In this paper, we propose an approach using an architecture suitable for 6G, which overcomes some of these issues by removing extra content from the packets, when network conditions are restricted.

Multipath TCP (MPTCP) is a protocol that enables the use of more than one subflow under the same TCP connection, which provides increased throughput due to the aggregated bandwidth. Therefore, the use of MPTCP can be very beneficial to video streaming applications since bandwidth is the most crucial parameter that improves the performance of such applications. However, the characteristics of the subflow paths might have a significant impact on application performance. In this study, we propose a path selection approach for MPTCP subflows in order to maximize the Quality of Experience (QoE) for adaptive HTTP streaming systems, which is currently one of the most popular video streaming application techniques. The paths are selected on the network layer by taking the bandwidth and delay differences into account. The disjointness of the paths is also considered in the proposed path selection approach. A genetic algorithm-based approach is utilized for the selection of the paths. Furthermore, we indicate how the characteristics of the paths affect QoE. The experimental results indicate that considering bandwidth and delay difference parameters jointly helps increase all QoE metrics, compared to the approaches that select paths considering either bandwidth or delay difference.

The emergence of a number of network communication facilities such as Network Function Virtualization (NFV), Software Defined Networking (SDN), the Internet of Things (IoT), Unmanned Aerial Vehicles (UAV), and in-network packet processing, holds a potential to meet the low latency, high precision requirements of various future multimedia applications. However, this raises the corresponding issues of how all of these elements can be used together in future networking environments, including newly developed protocols and techniques. This paper describes the architecture of an end-to-end video streaming platform for video surveillance, consisting of a UAV network domain, an edge server implementing in-network packet trimming operations with the use of Big Packet Protocol (BPP), utilization of Scalable Video Coding (SVC) and multiple video clients which connect to a network managed by an SDN controller. A Virtualized Edge Function at the drone edge utilizes SVC and in communication with the Drone Control Unit to manage the transmitted video quality. Experimental results show the potential that future multimedia applications can achieve the required high precision with the use of future network components and the consideration of their interactions.

This paper describes the effects of the Packet Wash process on the transmission of layered SVC video streams. We show how the packet size is adapted when using a number of different packing strategies, that map the video data into the BPP packets, and discuss the relationship between the packing strategies on the sender side and the chunk removal in the washing process. We demonstrate how the packing strategy causes different impacts on the number of and the sizes of the washed chunks. As the bandwidth reduces, more of the chunks in a packet get washed away. Although the receiver gets packets that are much smaller than those transmitted by the sender, it is still able to play video with a high QoE as zero packets are dropped. This traffic engineering enables a direct implementation of an in-network video adaption scheme. The experimental evaluation highlights that the effects of Packet Wash become more obvious in environments where there is limited bandwidth.

Big Packet Protocol (BPP), which is part of New IP, was designed to transfer packets for future networking applications, and aims to overcome obstacles within current networks for high precision services. One of the most important advantages of New IP is that it allows changes to packets during transmission. The strategy of BPP is to reduce the packet size by eliminating specific chunks, cutting out segments from the transferred video, rather than dropping or retransmitting packets. This provides an effective mechanism to enhance the performance of video streaming applications, by obtaining continuous delivery and minimum guaranteed quality at the receiver. In order to make video transmission over BPP effective, we need to select a video codec that can do multiple encodings for the same region, such as scalable video coding (SVC). To support such functionality, we have augmented the BPP packet structure in order to transfer video data. This paper describes the use of BPP for carrying video from servers to clients, and defines the packet structure for this purpose, plus the extensions needed to support SVC encoded video. To evaluate the proposed approach, we use SDN to facilitate BPP operations, with results showing a successful implementation of a system using these combined techniques.

The essence of this work is to show how SVC Scalable Video can be adaptated in the network in an effective way, when the Big Packet Protocol (BPP) is used. This demo shows the advantages of BPP, which is a recently proposed transport protocol devised for real-time applications. We will show that in-network adaption can be provided using this new protocol. We show how a network node can change the packets during their transmission, but still present a very usable video stream to the client. The preliminary results show that BPP is a good alternative transport for video transmission.

This paper describes the use of the Big Packet Protocol (BPP) for carrying video from servers to clients, and how SDN controllers can effectively manage the flow-rate and QoE, based on the available bandwidth. BPP relies on meta-data being injected into packets in order to provide information for network nodes on how to process those packets. Given specific commands, the network node can drop parts of the payload, called chunks in BPP. When using BPP, the strategy is not to drop whole packets, but to reduce the packet size be eliminating specific chunks. The approach allows for reducing the load on the network, when there is a limited bandwidth, by having a flow of packets regularly arriving at the receiver, so there is continuous delivery and minimum guaranteed quality. To make video transmission over BPP effective, a video encoder and decoder that can do multiple encodings for the same region is selected–namely scalable video coding (SVC). The results show the successful implementation of a system using these combined techniques.

Dynamic Adaptive Streaming over HTTP (DASH) is a technology designed to deliver video to the end-users in the most efficient way possible by providing the users to adapt their quality during streaming. In DASH architecture, the original content encoded into video streams in different qualities. As a protocol running over HTTP, the caches play an important role in DASH environment. Utilizing the cache capacity in these systems is an important problem where there are more than one encoded video files generated for each video content. In this paper, we propose a caching approach for DASH systems by predicting the future qualities of DASH clients. For the prediction, we use learning model, and the qualities that will be cached are determined by using this model. The learning model is designed using Recurrent Neural Networks (RNNs) and also Long Short Term Memory (LSTM) which is a special type of RNNs with default behavior of remembering information for long periods of time. We also utilize SDN technology to get some of the outputs for the learning algorithm. The simulation results show that predicting future qualities helps to reduce the underruns of the clients when cache storage is utilized.

Nowadays, video streaming over HTTP is one of the most dominant Internet applications, using adaptive video techniques. Network assisted approaches have been proposed and are being standardized in order to provide high QoE for the end-users of such applications. SAND is a recent MPEG standard where DASH Aware Network Elements (DANEs) are introduced for this purpose. As web-caches are one of the main components of the SAND architecture, the location and the connectivity of these web-caches plays an important role in the user's QoE. The nature of SAND and DANE provides a good foundation for software controlled virtualized DASH environments, and in this paper, we propose a cache location algorithm and a cache migration algorithm for virtualized SAND deployments. The optimal locations for the virtualized DANEs is determined by an SDN controller and migrates it based on gathered statistics. The performance of the resulting system shows that, when SDN and NFV technologies are leveraged in such systems, software controlled virtualized approaches can provide an increase in QoE.

Today, the most of the video streaming system provides quality adaptation and prefers to send their packets over HTTP. MPEG group has standardized Dynamic Adaptive HTTP Streaming (DASH) regarding this tendency on the adaptive HTTP streaming. Besides providing quality adaptation with a non-scalable codec, DASH standard also allows Scalable Video Coding (SVC) to adapt quality. Software Defined Networks (SDN) is a recently emerged networking paradigm. SDN enables to separate control and data plane of computer networks and hence provides flexibility to network operators to implement their own routing approaches. In this paper, we propose a system for increasing Quality of Experience (QoE) of SVC-DASH clients by utilizing SDN. Our experimental results show that the proposed system provides an increase in received video quality and decrease in outage duration and startup delay when compared to the performance of the client running over todays Internet implementing shortest path routing.

Abstract In peer‐to‐peer (P2P) video streaming systems, one of the most challenging parts is to schedule video data dissemination, that is, each peer should carefully select the partner(s) it receives video from and the partner(s) it sends data to. We believe that an agent‐based partner selection approach may improve the quality of streaming by taking both autonomy and dynamic plan selection into account in a goal‐oriented manner. In this study, a Belief‐Desire‐Intention agent architecture for partner selection in P2P video streaming systems is introduced. The major concern of our study is to exhibit how to select the best partner during video streaming session whilst maximizing the quality of video and minimizing delay and hop count. The effects and comparative results of executing proposed agent behaviours are evaluated in the study. The proposed autonomous agent‐based approach also provides an infrastructure in which the best plan for the achievement of optimum streaming goal can be dynamically determined and executed at runtime. Experimental results of the implementation have revealed to us that both of the partner selection methods (with or without agents) manage to increase the video quality. However, the agent‐based approach performs better in terms of received bitrate, delay and hop count during streaming.

Peer-to-peer (P2P) video streaming is a promising method for multimedia distribution over the Internet, yet many problems remain to be solved such as providing the best quality of service to each peer in proportion to its available resources, low-delay, and fault tolerance. In this paper, we propose a new bandwidth-aware multiple multicast tree formation procedure built on top of a hierarchical cluster based P2P overlay architecture for scalable video (SVC) streaming. The tree formation procedure considers number of sources, SVC layer rates available at each source, as well as delay and available bandwidth over links in an attempt to maximize the quality of received video at each peer. Simulations are performed on NS2 with 500 nodes to demonstrate that the overall performance of the system in terms of average received video quality of all peers is significantly better if peers with higher available bandwidth are placed higher up in the trees and peers with lower bandwidth are near the leaves.