PriCLeSS International Workshop, Monday Sept. 9 2024, at Inria Rennes
supported by Inria IRISA, the University or Rennes and Cominlabs
📣 Programme
9:00 Opening
-
Abstract: This talk presents a real-time implementation of a platform jointly developed by InterDigital and Philips that showcases use cases leveraging the MPEG volumetric codecs (MPEG-I V3C). We will detail how our platform enables interoperability within existing and emerging XR ecosystems, adaptive streaming, and real-time interactive playback of volumetric video on current and future client devices, for use in applications like telelearning, free-viewpoint sport replays, and 3D telepresence in connected ecosystems like the metaverse. MPEG’s Visual Volumetric Video-based Coding (V3C) standard is an extensive framework for the coding of volumetric video, from dynamic point clouds (V-PCC) to multi-view plus depth and multi-plane image (MIV) representations, to offer a single bitstream structure with a uniform bridge to systems-level standards. The V3C carriage standard defines how volumetric content can be stored, transported, and delivered to the end-user, and it repurposes existing 2D video hardware decoding capabilities and GPUs to decode and render volumetric video. We will present the architecture of InterDigital and Philips’ collaborative multi-platform implementation, which leverages Unity 3D, and is comprised of the V3C, HEVC and VVC decoders and rendering pipelines. We will evaluate the real-time performance on self-captured and MPEG-provided content to demonstrate that this multi-platform implementation enables efficient streaming of volumetric and 2D video to phones, tablets, and head-mounted displays, and is approaching readiness for distribution through cloud infrastructure.
-
Nicolas Mollet & Jean-François Vial (Imverse) Real-time hologram streaming (short talk)
Abstract: Imverse’s HoloLive software captures, streams and renders volumetric videos in real time – authentic 3D holograms, not avatars: one or up to ten 2D cameras with depth sensors (Microsoft Azure Kinect, Intel RealSense, Apple iPhone, etc.) for capture, a voxel-based engine for fast and performant data fusion and rendering, and a modular and scalable architecture for possible distributed deployment in the cloud and multi device interactive rendering. One key aspect – still under improvement – is the streaming itself, because the large amount of data implied by point-cloud or voxels needs efficient compression.
-
Adrien Gégout (Blacknut): Containerizing Cloud Gaming pipeline for an efficient Cloud Gaming infrastructure (short talk)
Abstract: This talk will outline the planned work on cloud gaming processes containerization and resource sharing, as part of a new joint thesis between Blacknut and WIDE Inria.
Video games are extremely heterogeneous processes, traditionally consumed alone on a single system, using graphical processing and generally very sensitive to resource access latency. For cloud gaming, this leads today to over scheduling of resources, wasting a part of them. This work plans to allow the collocation of several cloud gaming pipelines (game inputs reception and injection, game process execution and game video streaming) on the same virtual machine. Its main goal is to be able to finely orchestrate cloud gaming resources consumption.
Blacknut is a pure-player company, offering a cloud gaming service in several countries in the world. The company offers a subscription-based product that includes the infrastructure for streaming video games. Concerned about its impact and resource consumption, Blacknut strives for continuous improvement and seeks to optimize its service and infrastructure, trying to lead the way to an efficient cloud gaming infrastructure.
10:45 ☕️ Coffee Break
-
Rufael Mekuria (Huawei): Standards for streaming at scale
For content owners and streamers, a key challenge remains: how do I reach large audiences cost effectively?
In this presentation we give an overview of media standards that enable streaming at scale.
We provide a quick introduction to HTTP streaming as it is used today and the common content delivery architecture.
Two key changes in this content delivery architecture include moving from push to pull based delivery
and moving from MPEG-2 transport stream to ISO Base Media File Format based delivery.We provide details on using the ISO Base media file format for text, video, audio and metadata and
present what a segmented ISO BMFF file looks like. Also we show how the downloading of content
initiated by the client introduces the problem of bandwidth estimation at the client. We provide a
quick overview of current rate adaptation solutions as adopted in dash.js, DVB-DASH and the DASH-IF IOP.Next, we provide more details about the scope and standardization of MPEG-DASH and how profiles have
been developed to cover different industry needs and player requirements. Next we discuss how these
profiles and requirements eventually led to the Common Media Application Format.Next we present some practical and deployment architecture choices, such as using just-in-time versus static packaging for live and VoD workflows.
We present the DASH-IF live ingest protocol and detail the emerging MPEG-DASH Redundant encoding and packaging specification (REaP)
to enable efficient deployment of live workflows with dual encoding/packaging.Next, we briefly introduce some standards that enable delivery at scale, by making use of specific facilities for caching or streaming inside the ISP network.
In addition we briefly mention content steering and the DVB-DASH mechanisms for multi-CDN delivery, and
the tradeoff between server side versus client side steering. -
Abstract: In the pursuit of greener video streaming, the industry confronts the pressing issues of electricity consumption and carbon emissions. Current multi-CDN strategies prioritize QoE, mostly neglecting evolving CDN requirements, including their demand for a better power efficiency. CDNs lack of mechanisms to communicate their evolving requirements, hindering the optimization of multi-CDN balancing algorithms. This talk explores strategies to establish a harmonious content steering service in multi-CDN world, emphasizing the strategic exposure of internal metrics like power efficiency and profitability, enabling CDNs to align their traffic loads with their best interests while contributing to a greener and more sustainable streaming future.
12:25 🍽 Lunch
-
Amélie Gonzalez (IRISA): Should you (re)write your network drivers in Rust? Takeaways of Developing a Rust UDP Tunneling Driver in Linux (short talk)
Abstract: Kernel optimization encompasses both drivers developed for particular pieces of hardware, as well as core kernel subsystems that expose an API for those drivers. In the realm of networking, that is `net`, its network stack, and the set of APIs that enable driver developers to plug their own drivers into the stack. Of course, network drivers are subject to the same scrutiny and micro-optimizations as all other drivers, and the potential those latter modifications have to introduce errors, particularly memory management errors, is a cause for concern. Introducing memory management bugs into network drivers can have catastrophic consequences, but, on the other hand, being overly cautious could result in wasting very precious CPU cycles. Following recent trends regarding the adoption of Rust for systems programming, a group of developers founded the Rust for Linux project, aiming to integrate support for the Rust programming language into the Linux kernel. Rust promises performance on par with C with the benefits of memory protection verified at compile time, as well as a modern type system providing OO-like programming with structures, traits, generics, and such. In this presentation, I will highlight the challenges and takeaways of developing a UDP tunneling driver inspired by WireGuard in pure Rust in the Linux kernel, with a focus on how the language and its type system guided and helped us while trying to interface with the existing C API. We aimed to, and almost entirely succeeded in, creating a driver with no trusted code. Said driver, while not production-ready, is able to carry connections bidirectionally like other layer-3 tunneling systems would. Our benchmarks reveal that our Rust driver can perform at a similar level to an equivalent C driver on a gigabit link but with a slight cost in performance for both latency and throughput. Potential causes for that loss are discussed both in terms of currently known limitations of Rust for Linux’s build system, as well as limitations in our driver.
-
Honoré Césaire Mounah (Inria): Understanding Wireguard behavior at scale (short talk)
Abstract: Today, Virtual Private Networks (VPN) are used everywhere. They are widely used both in business and for personal use. In 2017, Jason Donenfeld proposed a new VPN protocol, Wireguard, designed to be lighter and faster than existing protocols, while ensuring a high level of security. His evaluations show that Wireguard outperforms OpenVPN and IKEv2/IPSec in terms of network throughput. On the other hand, several studies have formally demonstrated the security of Wireguard’s cryptographic protocols. However, not much work exists to evaluate Wireguard’s ability to scale, which is important given the growing number of VPN users. Our work is to evaluate its ability to support a large number of clients and to scale. On a server with 18 cores and a 25 Gbps network card, we study the network throughput and network latency of several Wireguard implementations in different network configurations. Our analysis shows scalability problems at various levels, notably the distribution of network interrupts over multiple CPU cores, and the lack of load-balancing mechanism for network functions. Throughput peaks at 2.9 Gbps, and latency is 14 ms. We show that with a few configurations at the OS level, we can improve Wireguard’s scalability. Specifically, we improve throughput by 8x to 21 Gbps and reduce client latency to 1 ms, corresponding to a 14x improvement.
15:30 ☕️ Coffee Break
-
Juncheng Yang (CMU): FIFO queues are all you need for cache eviction
Abstract: As a cache eviction algorithm, FIFO has a lot of attractive properties, such as simplicity, speed, scalability, and flashfriendliness. The most prominent criticism of FIFO is its low efficiency (high miss ratio). In this work, we demonstrate a simple, scalable FIFObased algorithm with three static queues (S3-FIFO). Evaluated on 6594 cache traces from 14 datasets, we show that S3- FIFO has lower miss ratios than state-of-the-art algorithms across traces. Moreover, S3-FIFO’s efficiency is robust — it has the lowest mean miss ratio on 10 of the 14 datasets. FIFO queues enable S3-FIFO to achieve good scalability with 6× higher throughput compared to optimized LRU at 16 threads. Our insight is that most objects in skewed workloads will only be accessed once in a short window, so it is critical to evict them early (also called quick demotion). The key of S3-FIFO is a small FIFO queue that filters out most objects from entering the main cache, which provides a guaranteed demotion speed and high demotion precision.
-
Kévin Vermeulen (LAAS-CNRS): The best of both worlds: high availability CDN routing without compromising control
Abstract:
Content delivery networks (CDNs) provide fast service to clients by replicating content at geographically distributed sites. Most CDNs route clients to a particular site using anycast or unicast with DNS-based redirection. We analyze anycast and unicast and explain why neither of them provides both precise control of user-to-site mapping and high availability in the face of failures, two fundamental goals of CDNs. Anycast compromises control (and hence performance), and unicast compromises availability. We then present new hybrid techniques and demonstrate via experiments on the real Internet that these techniques provide both a high level of traffic control and fast failover following site failures. -
Alexandre Duvivier (Broadpeak): How to run an HTTPS server at 700 Gbps on your typical dual socket server (short talk)
Abstract: In this presentation, we explain how nowadays caching servers can reach high HTTP performances using the zerocopy mechanism. We also show that using HTTPS reduces drastically the gains from this mechanism and requires adaptations to fully benefit from it. Finally, we present the results on our custom caching server, which reaches performances of around 700Gbps in HTTPS.
17:45 Wrap-up
✒️ Registration
Registration is now closed. See you on Tuesday!
🗺 Venue
The workshop will be taking place at the Centre de conférence Inria, on the Beaulieu Campus of the University of Rennes 1, Avenue du Général Leclerc 35042 Rennes Cedex. More detailed travel information can be found here.
Previous WOS Editions
