WOS10: Inria Workshop On Systems
Oct. 12th 2021, at Inria Rennes and online
9h00 – Welcome
9h15 – Session 1 – Energy/frugal computing
Anne-Cécile Orgerie (Univ Rennes, CNRS, Inria, IRISA): Measuring and modeling the energy consumption of Cloud infrastructures
Abstract: Cloud computing is increasingly spanning worldwide, with digital services hosted all around the globe and often belonging to complex systems, utilizing many other services and hardware resources themselves. Along with this increase comes an alarming growth of Cloud devices and their related energy consumption. Despite the Cloud systems’ complexity, understanding how they consume energy is important in order to hunt wasted Joules. This talk will deal with measuring the energy consumption of Cloud infrastructures, deriving models from these measurements and implementing these models into simulation tools that can be used to experiment new energy-efficient strategies.
Romain Rouvoy (Université de Lille, Inria): How green is your cloud: from measurements to actionable insights
Abstract: The cloud is often singled out as one of the major culprits in the explosion of digital environmental impact worldwide. While the success of this paradigm over the years is now undeniable, referring to a single environmental indicator, such as PUE, does not allow for an in-depth analysis of the strengths and weaknesses of a public or private cloud operator. This talk will therefore focus on exploring the methods and tools that can effectively reduce the environmental impact of the cloud without necessarily constraining its usages.
10h45 – Break
11h15 – Session 2 – Network inference, analysis of network data
Thibault Maho (Inria Rennes): SurFree: a fast surrogate-free black-box attack
Abstract: Machine learning classifiers are critically prone to evasion attacks. Adversarial examples are slightly modified inputs that are then misclassified, while remaining perceptively close to their originals. Last couple of years have witnessed a striking decrease in the amount of queries a black box attack submits to the target classifier, in order to forge adversarials. This particularly concerns the black-box score-based setup, where the attacker has access to top predicted probabilites: the amount of queries went from to millions of to less than a thousand.
This talk presents SurFree, a geometrical approach that achieves a similar drastic reduction in the amount of queries in the hardest setup: black box decision-based attacks (only the top-1 label is available). We first highlight that the most recent attacks in that setup, HSJA, QEBA and GeoDA all perform costly gradient surrogate estimations. SurFree proposes to bypass these, by instead focusing on careful trials along diverse directions, guided by precise indications of geometrical properties of the classifier decision boundaries. We motivate this geometric approach before performing a head-to-head comparison with previous attacks with the amount of queries as a first class citizen. We exhibit a faster distortion decay under low query amounts (few hundreds to a thousand), while remaining competitive at higher query budgets.
Chadi Barakat (Inria Sofia Antipolis): Bridging the gap between network measurements and quality of experience: the video streaming case
Abstract: We overview in this talk the set of our activities on the experimentation, measurement, and modeling of Quality of Experience for video streaming traffic. Video streaming is one of the services posing a serious challenge for network operaters and content providers as for its popularity, its greediness in terms of network resources, and its sensitivity to the level of service provided by the network. The management of video streaming is thus a challenging task for operators, and this challenge is even further increased with the shift of video streaming towards end-to-end encryption. In this project, we have worked on the characterization of video streaming traffic and the building of predictive models for the quality of experience (QoE) of end users that allow operators to take appropriate network management decisions. We will start by overviewing our experimental framework built around intelligent controlled experimentation, then move to present a list of results about the analysis of video streaming traffic and the associated level of quality of experience. We will discuss about our modeling work for the prediction of QoE using machine learning and network measurements that are either collected out-of-band before the arrival of the video streaming traffic (the prediction case) or in-band from the encrypted video traffic itself (the inference case). We will also discuss the effect of the viewing port resolution on the Quality of Experience and the video traffic itself, and propose machine learning models that can succefully infer the viewport class (either SD or HD), and with a lower precision its exact resolution. In the meantime, we present results on the respect of the video player for the viewport resolution and the waste of bandwidth that can occur.
12h45 – Lunch Break. On your own: Be aware that lunch cannot be provided for sanitary reasons.
14h30 – Session 3 – Performance monitoring
Jalil Boukhobza (ENSTA Bretagne), François Trahay (Télécom SudParis / Institut Polytechnique): EZIOTracer: unifying kernel and user space I/O tracing for data-intensive applications
Abstract: Tracing is a popular method for evaluating, investigating, and modeling the performance of today’s storage systems. Tracing has become crucial with the increase in complexity of modern storage applications/systems, that are manipulating an ever-increasing amount of data and are subject to extreme performance requirements. There exists many tracing tools focusing either on the user-level or the kernel-level, however we observe the lack of a unified tracer targeting both levels: this prevents a comprehensive understanding of modern applications’ storage performance profiles. In this paper, we present EZIOTracer, a unified I/O tracer for both (Linux) kernel and user spaces, targeting data intensive applications. EZIOTracer is composed of a userland as well as a kernel space tracer, complemented with a trace analysis framework able to merge the output of the two tracers, and in particular to relate user-level events to kernel-level ones, and vice-versa. On the kernel side, EZIOTracer relies on eBPF to offer safe, low-overhead, low memory footprint, and flexible tracing capabilities. We demonstrate using FIO benchmark the ability of EZIOTracer to track down I/O performance issues by relating events recorded at both the kernel and user levels. We show that this can be achieved with a relatively low overhead that ranges from 2% to 26% depending on the I/O intensity.
Nominoe Kervadec (Broadpeak): Building a high-performance benchmarking and load-testing tool for video streaming and content delivery systems
Abstract: QoE of users can be impacted by errors or undue latency from streaming servers. Thus, as the streaming server must transfer large amount of data without any hiccups, it must be optimized for reliable and consistently high performance. In order to assess its performance or detect issues, it is critical to generate as realistic as possible workloads, reproducing subtle timing patterns such as implicit synchronization of players, and avoiding pitfalls of synthetic workloads. To this end, at Broadpeak, we’ve implemented a high-performance benchmarking tool able to emulate precisely HLS and DASH players while leveraging the Chromium engine for realism. We describe alternatives for video streaming benchmarking, and present some key aspects of our tool. We conclude by giving a glimpse of its capabilities and uses so far.
Freysteinn Alfredsson (Karlstad University, Sweden): Bringing packet queueing to XDP
Abstract: The Linux eXpress Data Path, or XDP, has found numerous uses in the industry,
such as DoS attack mitigation, load-balancers, and intrusion prevention systems.
XDP provides a high-performance programmable network data path using the eBPF
framework and allows programmers to process packets early out of the driver.
While XDP excels in forwarding packets, it currently has no mechanism for
queueing or reordering packets and cannot implement traffic scheduling policies.
In this talk, we present our ongoing work to address this challenge. We intend
to design a programmable packet scheduling extension for the XDP framework using
recently proposed schemes for programmable queues. This extension allows
programmers to define their packet schedulers using eBPF while benefiting from
the XDP fast data path.
Yohann Ghigoff (Orange): BMC: Accelerating Memcached using Safe In-kernel Caching and Pre-stack Processing
Abstract: In-memory key-value stores are critical components that help scale large internet services by providing low-latency access to popular data. Memcached, one of the most popular key-value stores, suffers from performance limitations inherent to the Linux networking stack and fails to achieve high performance when using high-speed network interfaces. While the Linux network stack can be bypassed using DPDK based solutions, such approaches require a complete redesign of the software stack and induce high CPU utilization even when client load is low.
To overcome these limitations, we present BMC, an in-kernel cache for Memcached that serves requests before the execution of the standard network stack. Requests to the BMC cache are treated as part of the NIC interrupts, which allows performance to scale with the number of cores serving the NIC queues. To ensure safety, BMC is implemented using eBPF. Despite the safety constraints of eBPF, we show that it is possible to implement a complex cache service. Because BMC runs on commodity hardware and requires modification of neither the Linux kernel nor the Memcached application, it can be widely deployed on existing systems. BMC optimizes the processing time of Facebook-like small-size requests. On this target workload, our evaluations show that BMC improves throughput by up to 18x compared to the vanilla Memcached application and up to 6x compared to an optimized version of Memcached that uses the SO_REUSEPORT socket flag. In addition, our results also show that BMC has negligible overhead and does not deteriorate throughput when treating non-target workloads.
16h30 – Coffee Break
17h00 – Session 4 – Identity and Personal Data (1h10)
Nabil Ghanmi (ARIADNEXT): A full remote identity verification system for a secure KYC procedure
Abstract: In an economic environment that is increasingly being digitalized, it is becoming essential for companies to rethink their customer journey in order to offer them fast, reliable and ergonomic online services. This affects more particularly the customer onboarding stage, which must bring mutual trust between all the players. In this talk, I will present a fully automated onboarding system including various KYC services such as: identity document verification, face recognition and liveness detection. This system provides the confidence needed at the initial contact stage and helps combat fraud and meets regulatory requirements.
From a freely captured image (using flat scanner, smartphone or any other device), the identity document is firstly identified. Then, personal information is extracted and analyzed based on advanced ML techniques. Several automated verifications are also performed to check the document authenticity. Furthermore, in order to ensure that the user wishing to authenticate is who he claims to be behind the screen, biometric tools are used to invite him to take a selfie or a facial recognition video and ask him to perform a series of movements which are then analyzed and cross-referenced with the identity document data. These biometric analysis are performed using recent deep learning systems that have proven to be very effective and robust to various presentation attack cases. Once the analysis has been completed, a real-time verdict on the authenticity, validity and conformity of the analyzed document, as well as a control report on the holder’s identity are established.
Our system is used by major financial institutions in several European countries, many Fintechs, online gaming operators as well as many service providers needing confidence in an increasingly digital and regulated world.
Vincent Roca (Inria Grenoble): TousAntiCovid: closeup on the contact and presence digital tracing features of the French app
Abstract: The TousAntiCovid French app now features two digital tracing functionalities: contact tracing (since June 2020) through the ROBERT protocol, and presence tracing (since June 2021) through the CLÉA protocol. In this talk we will introduce these two complementary systems: the motivations, the technical principles and associated challenges, how it compares to the alternatives, and some statistics. More details on: https://github.com/ROBERT-proximity-tracing/documents and https://gitlab.inria.fr/stopcovid19/CLEA-exposure-verification