Title: DRLM: Deep Reinforcement Learning-Based LLM Query Orchestration in Edge Environments

Authors: Reza Farahani (TU Wien, Austria), Zoha Azimi (AAU, Austria), Mario Colosi (University of Messina, Italy), Lauri Lov\’en (University of Oulu, Finland), Christian Timmerer (AAU, Austria), Schahram Dustdar (TU Wien, Austria)

Venue: IEEE Global Communications Conference (GLOBECOM), Macau S.A.R., China, 7 – 11 December 2026

Abstract:

Large language model (LLM) services increasingly process heterogeneous queries with diverse latency, accuracy, and resource requirements. While edge deployment reduces response time, the heterogeneity of devices and the diversity of model families, parameter scales, and quantization levels make efficient LLM query orchestration challenging. This paper introduces DRLM, a Deep Reinforcement Learning-based LLM query orchestration framework in edge environments. DRLM integrates two lightweight predictors: (i) a class-conditioned quality estimator that maps queries to semantic categories and infers model performance, and (ii) a feature-driven latency predictor that estimates inference time across model-device configurations. These predictions, combined with system state (resource utilization and queue dynamics), feed a factorized Proximal Policy Optimization (PPO) agent that performs state-aware orchestration decisions. To enable data-driven orchestration, we construct a large-scale benchmarking dataset with 223 835 measurements spanning 1258 queries, 6 query classes, 8 model families (32 deployed instances), 5 quantization levels, and heterogeneous edge devices. Evaluation on a realistic 64-node edge cluster and comparison with three baselines and two state-of-the-art methods show that DRLM reduces inference latency by up to 51 % and queuing delay by up to 67 %, while incurring at most 8 % accuracy loss. DRLM further improves latency under increasing workloads up to 61.4 %, demonstrating robust and stable orchestration.

Content-Adaptive Encoding Pass Selection for Efficient Video Streaming

International Workshop on Multimedia Signal Processing (MMSP)

September 22 – September 24, 2026

Istanbul, Türkiye

[PDF]

Mohammad Ghasempour (AAU, Austria), Hadi Amirpour (AAU, Austria), Christian Timmerer (AAU, Austria)

Abstract: As video streaming continues to grow in scale, improving the efficiency of video encoding has become increasingly important to balance visual quality and computational cost. Multi-pass encoding is widely adopted to enhance compression efficiency, improve rate control, and achieve more stable quality by leveraging additional analysis of video content prior to encoding. These benefits come with the cost of increased computational complexity. In this paper, we show that the benefits of multi-pass encoding vary substantially across video content and encoding configurations in adaptive video streaming. Motivated by this observation, we propose the Adaptive Encoding Pass Selection (AEPS), a lightweight content-adaptive framework that estimates the benefits of multi-pass prior to encoding and enables selective use of single-pass encoding to reduce encoding time. Experimental results demonstrate that the AEPS framework substantially reduces encoding time while maintaining compression performance and stability, achieving an average 25.3% reduction in encoding time with only a 2.23% increase in bitrate. We show that the preprocessing and decision-making overhead of AEPS is approximately 1155 times lower than the time required for multi-pass encoding in adaptive streaming.

Title: Towards AI-Assisted Sustainable Adaptive Video Streaming Systems: Tutorial and Survey

Authors: Reza Farahani, Zoha Azimi, Christian Timmerer, Radu Prodan

Journal: ACM Computing Surveys

Abstract: Improvements in networking technologies and the steadily increasing number of users, as well as the shift from traditional broadcasting to streaming content over the Internet, have made video applications (Video-on-Demand (VoD) and live streaming) predominant sources of traffic. Recent advances in Artificial Intelligence (AI) and its widespread application in various academic and industrial fields have focused on designing and implementing a variety of video compression and content delivery techniques to improve user Quality of Experience (QoE). However, providing high QoE services results in increased energy consumption and a larger carbon footprint across the service delivery path, extending from the end-user’s device through the
network and service infrastructure (e.g., cloud providers). Despite the importance of energy efficiency in video streaming, there is a lack of comprehensive surveys covering state-of-the-art AI techniques and their applications throughout the video streaming lifecycle. Existing surveys typically focus on specific parts, such as video encoding, delivery networks, playback, or quality assessment, without providing a holistic view of the entire lifecycle and its impact on energy consumption and QoE. Motivated by this research gap, this article provides a comprehensive overview of the video streaming lifecycle, content delivery, energy, and Video Quality Assessment (VQA) metrics and models, and AI techniques employed in video streaming. In addition, it conducts an in-depth state-of-the-art analysis of AI-driven approaches for improving the energy efficiency of end-to-end video streaming systems across encoding, delivery, playback, and VQA stages. It further discusses key challenges in AI-assisted streaming, including ethical concerns (privacy, bias, security), deployment barriers (dataset limitations, scalability, inference efficiency), protocol-level latency-computation trade-offs (e.g., QUIC and WebRTC), and the energy implications of Generative AI and semantic streaming.

Selective Multi-Pass Encoding for Cost-Efficient Video Streaming

International Broadcasting Convention (IBC)

[PDF]

Mohammad Ghasempour (AAU, Austria), Hadi Amirpour (AAU, Austria), and Christian Timmerer (AAU, Austria)

Abstract: As video streaming scales across platforms, resolutions, and devices, encoding efficiency has become critical to maintaining quality while controlling computational cost and energy consumption. Multi-pass encoding is widely used in streaming workflows to improve compression efficiency, rate-control accuracy, and quality consistency. However, its computational overhead is applied uniformly across all content, even when additional passes deliver minimal benefit. At scale, this results in unnecessary processing, higher computational cost, and increased energy consumption. This paper presents a content-adaptive strategy, CASE, that predicts whether additional encoding passes would provide meaningful gains using a lightweight mechanism that derives spatial and temporal features from each video segment. We evaluated the approach in two production-oriented scenarios using local and cloud-based video encoders. Results show that the method reduces computational time and encoding cost, with minimal impact on compression efficiency and visual quality. Experimental results show that CASE reduces encoding time by 25.3% on average with only a 2.23% bitrate increase, while its preprocessing and decision overhead is about 1155 times lower than multi-pass encoding time.