With the emergence of containerization 10 years ago, we saw a compact, convenient and portable way of running apps directly concurrently with virtualization. The major difference is in the architecture. Containers share the same kernel as the guest and then do not virtualize low-layer components like the Central Processing Unit (CPU). On the one hand, they are lighter and more flexible than virtual machines (VMs). On the other hand, VMs can more precisely meet the low-layer needs and are completely autonomous systems. Nowadays, what is the best architecture to use to develop an application? In this paper, we will study the two main virtual methods of deploying this. We will compare both methods on several criteria: compatibility based on user experience and the ease of installation/deployment, scalability based on the automatic elasticity facing the workload and energy efficiency in terms of energy and computer resources. After the tests, we conclude that the containerization option is the most ecologically advantageous option in terms of energy consumption.
Nowadays, the majority of media Internet traffic consists of H.264-encoded streaming videos due to its high compatibility. One of the most popular streaming technology used to deliver videos over Internet is Dynamic Adaptive Streaming over HTTP (DASH). It transmit the video as a sequence of independent short video segments tailored to the receiver’s limitations (related to several factors such as available bandwidth or resolution on reception), aiming to enhance the Quality of Experience. In this paper, we present two types of datasets created from 4065 video segments of 2 seconds. The first type consists of extracting features related to color, space, and time from the segments across different vertical resolutions (240, 360, 480, 720, 1080, 1440 and 4K). The second one includes several quality metrics obtained from the same segments when they are encoded with different compression levels in the different resolutions.
In the contemporary digital media environment, video encoding is of paramount importance for applications such as streaming and video conferencing. With the increasing demand for higher resolutions and immersive video, optimizing videos for Internet consumption necessitates the use of the latest codecs. Additionally, live event broadcasts require fast and efficient codecs, making GPU codecs essential. This paper analyses the coding efficiency of video encoders on CPUs and GPUs with 4K content. We have selected three families of encoders (H264/AVC, HEVC, and AV1) and have computed the Bjøntegaard Delta (BD) using a set of 15UHDvideos to compare the results of the CPU implementation (libx264, libx265, and librav1e) with their GPU version (h264_nvenc, h265_nvenc, and av1_nvenc) of each family. Besides, we have measured the mean encoding time per frame (as the test videos have different duration). The BD rate analysis shows that the CPU implementations of the codecs for H264 and AV1 (libx264 and librav1e) require in mean more than 20% bitrate (for both metrics PSNR and VMAF) compared to the GPU implementations (h264_nvenc and av1_nvenc). However, the CPU implementation of the H265 (libx265) codec required around 7.5% less bitrate (for both metrics) than the GPU implementation (hevc_nvenc). The encoding times reveal that the mean encoding time per frame for the GPU encoders is very similar (109 ms, 108.39 ms, and 117.77 ms) and they achieve a great time reduction compared to the CPU version (1.3 times faster for H264, 2.35 times faster for HEVC, and 68.14 times faster for AV1). Therefore, the GPU implementations of these encoders reduce the encoding time drastically, specially in the case of AV1, and lead to a noticeable reduction of the bitrate (except for HEVC).
Multimedia streaming has become integral to modern living, reshaping entertainment consumption, information access, and global engagement. The ascent of cloud computing, particularly serverless architectures, plays a pivotal role in this transformation, offering dynamic resource allocation, parallel execution, and automatic scaling-especially beneficial in HTTP Adaptive Streaming (HAS) applications. This study presents an event-driven serverless cloud edge platform with graphics processing units (GPUs), managed by Knative, tailored for video encoding. Two implementations of the High Efficiency Video Coding (HEVC) codec have been encapsulated in the functions: HEVC NVENC (Nvidia Encoder), that uses GPU acceleration, and x265 that only uses CPUs. Experiments focused on measuring the impact of replica requested resources on cold start, scalability and resource consumption with different allocated resources on slim and fat virtual machines (VMs). The best results are obtained when four slim replicas of the functions are deployed on a fat VM with a 8.4% reduction in encoding time for x265 and a 15.2% improvement for HEVC NVENC compared with other deployment scenarios. Comparatively, HEVC NVENC encoding is 8.3 times faster than x265. In multiresolution scenarios, GPU encoding drastically reduces segment encoding time by a factor of 12.4 between non-GPU and GPU-accelerated. These findings are important for live streaming applications where low latency is critical at all stages of the streaming process.
10 April 2025 in Software: Practice and Experience
Background: Modern microservice architectures demand not only modularity, scalability, and maintainability but also adaptation to dynamic requirements. For applications deployed on Kubernetes, sidecar proxies have been used to separate the operational features (such as security, networking, or monitoring) from the application business logic by intercepting traffic to and from microservices or functions in serverless platforms. Existing proxy sidecar implementations such as Envoy offer these operational features as chains of logic that cannot be composed at deployment time.Objective: This work introduces K8Sidecar, which leverages the operator pattern to dynamically integrate a chain of proxy sidecar containers into microservices or serverless architectures deployed on top of Kubernetes.Methods: The study presents the architecture of the software and describes the Java and Go libraries provided to develop new proxy sidecars.Results: We provide illustrative examples and evaluate the impact of the number of injected sidecars on cold‐start and latency and compare them with Envoy.Conclusion: Results show that for up to 5 sidecars, the chain‐of‐sidecars approach (especially using the Go library) remains reasonably close to Envoy in terms of both cold start and latency.
Weed control in permanent crops remains a costly and environmentally demanding task. While UAV-based imaging has advanced rapidly, publicly available datasets for weed detection in orchard settings are scarce, particularly those including multispectral and multi-temporal acquisitions. This lack of resources limits the development of robust machine learning models for cross-spectral adaptation and seasonal generalization.