arXiv subjects

Mahdi Dolati

Publications and source records attributed to Mahdi Dolati.

4 recordsLinked to original sources

arXiv2026

Thin-Client Interactive Gaussian Adaptive Streaming over HTTP/3

Recent advancements in 3D Gaussian Splatting (3DGS) have enabled photorealistic rendering of complex scenes, yet widespread adoption on mobile and Extended Reality (XR) devices is hindered by substantial computational and bandwidth requirements. While existing solutions often focus on model compression for client-side rendering, they still demand significant GPU power, limiting applicability on resource-constrained hardware. We propose TIGAS (Thin-client Interactive Gaussian Adaptive Streaming), a remote rendering framework offloading rasterization to a backend. To bypass the prohibitive latencies connected to fluctuating network conditions, TIGAS streams view-dependent 2D projections to a lightweight web client over QUIC, minimizing head-of-line (HoL) blocking. A dedicated ABR algorithm adapts rendering quality to fluctuating network conditions, maintaining motion-to-photon latency within strict 6DoF interactive constraints. Furthermore, we discuss the integration of an experimental WebGPU super-resolution pipeline to analyze the trade-offs between perceptual quality enhancements and thin-client processing bottlenecks. We extensively evaluate TIGAS across multi-continental environments using 14 3DGS models and real 6DoF EyeNavGS movement traces. Powered by a backend rendering frames in under 10 milliseconds, TIGAS maintains latency within interactive thresholds while achieving an average SSIM of 0.88, serving both as a robust testbed for 3DGS streaming research and a capable delivery system. The source code is available at: https://github.com/Rekenar/GaussianAdaptiveStreamer.

eess.IV↗

arXiv2025

YTLive: A Dataset of Real-World YouTube Live Streaming Sessions

Live streaming plays a major role in today's digital platforms, supporting entertainment, education, social media, etc. However, research in this field is limited by the lack of large, publicly available datasets that capture real-time viewer behavior at scale. To address this gap, we introduce YTLive, a public dataset focused on YouTube Live. Collected through the YouTube Researcher Program over May and June 2024, YTLive includes more than 507000 records from 12156 live streams, tracking concurrent viewer counts at five-minute intervals along with precise broadcast durations. We describe the dataset design and collection process and present an initial analysis of temporal viewing patterns. Results show that viewer counts are higher and more stable on weekends, especially during afternoon hours. Shorter streams attract larger and more consistent audiences, while longer streams tend to grow slowly and exhibit greater variability. These insights have direct implications for adaptive streaming, resource allocation, and Quality of Experience (QoE) modeling. YTLive offers a timely, open resource to support reproducible research and system-level innovation in live streaming. The dataset is publicly available at github.

cs.MM↗

arXiv2025

GOGH: Correlation-Guided Orchestration of GPUs in Heterogeneous Clusters

The growing demand for computational resources in machine learning has made efficient resource allocation a critical challenge, especially in heterogeneous hardware clusters where devices vary in capability, age, and energy efficiency. Upgrading to the latest hardware is often infeasible, making sustainable use of existing, mixed-generation resources essential. In this paper, we propose a learning-based architecture for managing machine learning workloads in heterogeneous clusters. The system operates online, allocating resources to incoming training or inference requests while minimizing energy consumption and meeting performance requirements. It uses two neural networks: the first provides initial estimates of how well a new model will utilize different hardware types and how it will affect co-located models. An optimizer then allocates resources based on these estimates. After deployment, the system monitors real performance and uses this data to refine its predictions via a second neural network. This updated model improves estimates not only for the current hardware but also for hardware not initially allocated and for co-location scenarios not yet observed. The result is an adaptive, iterative approach that learns over time to make more effective resource allocation decisions in heterogeneous deep learning clusters.

cs.DC↗

arXiv2024

Coordinated Sampling in SDNs with Dynamic Flow Rates

Traffic sampling has become an indispensable tool in network management. While there exists a plethora of sampling systems, they generally assume flow rates are stable and predictable over a sampling period. Consequently, when deployed in networks with dynamic flow rates, some flows may be missed or under-sampled, while others are over-sampled. This paper presents the design and evaluation of dSamp, a network-wide sampling system capable of handling dynamic flow rates in Software-Defined Networks (SDNs). The key idea in dSamp is to consider flow rate fluctuations when deciding on which network switches and at what rate to sample each flow. To this end, we develop a general model for sampling allocation with dynamic flow rates, and then design an efficient approximate integer linear program called APX that can be used to compute sampling allocations even in large-scale networks. To show the efficacy of dSamp for network monitoring, we have implemented APX and several existing solutions in ns-3 and conducted extensive experiments using model-driven as well as trace-driven simulations. Our results indicate that, by considering dynamic flow rates, APX outperforms the existing solutions by up to 10% in sampling more flows at a given sampling rate.

cs.NI↗