The October 2021 issue

This October 2021 issue contains two technical papers, two educational contributions, and one editorial note.

When latency matters

Several emerging classes of interactive applications are demanding for extremely low-latency to be fully unleashed, with edge computing generally regarded as a key enabler thanks to reduced delays. This paper presents the outcome of a large-scale end-to-end measurement campaign focusing on task-offloading scenarios, showing that moving the computation closer to the end-users, alone, may turn out not to be enough. Indeed, the complexity associated with modern networks, both at the access and in the core, the behavior of the protocols at different levels of the stack, as well as the orchestration platforms used in data-centers hide a set of pitfalls potentially reverting the benefits introduced by low propagation delays. In short, we highlight how ensuring good QoS to latency-sensitive applications is definitely a multi-dimensional problem, requiring to cope with a great deal of customization and cooperation to get the best from the underlying network.

REDACT

Refraction networking is a promising censorship circumvention technique in which a participating router along the path to an innocuous destination deflects traffic to a covert site that is otherwise blocked by the censor. However, refraction networking faces major practical challenges due to performance issues and various attacks (e.g., routing-around-the-decoy and fingerprinting). Given that many sites are now hosted in the cloud, data centers offer an advantageous setting to implement refraction networking due to the physical proximity and similarity of hosted sites. We propose REDACT, a novel class of refraction networking solutions where the decoy router is a border router of a multi-tenant data center and the decoy and covert sites are tenants within the same data center. We highlight one specific example REDACT protocol, which leverages TLS session resumption to address the performance and implementation challenges in prior refraction networking protocols. REDACT also offers scope for other designs with different realistic use cases and assumptions.

An educational toolkit for teaching cloud computing

In an educational context, experimenting with a real cloud computing platform is very important to let students understand the core concepts, methodologies and technologies of cloud computing. However, API heterogeneity of cloud providers complicates the experimentation by forcing students to focus on the use of different APIs, and by hindering the jointly use of different platforms. In this paper, we present EasyCloud, a toolkit enabling the easy and effective use of different cloud platforms. In particular, we describe its features, architecture, scalability, and use in our cloud computing courses, as well as the pedagogical insights we learnt over the years.

Data-driven networking research

We in Google's various networking teams would like to increase our collaborations with academic researchers related to data-driven networking research. There are some significant constraints on our ability to directly share data, which are not always widely-understood in the academic community; this document provides a brief summary. We describe some models which can work - primarily, interns and visiting scientists working temporarily as employees, which simplifies the handling of some confidentiality and privacy issues. We describe some specific areas where we would welcome proposals to work within those models.

Machine learning-based analysis of COVID-19 pandemic impact on US research networks

This study explores how fallout from the changing public health policy around COVID-19 has changed how researchers access and process their science experiments. Using a combination of techniques from statistical analysis and machine learning, we conduct a retrospective analysis of historical network data for a period around the stay-at-home orders that took place in March 2020. Our analysis takes data from the entire ESnet infrastructure to explore DOE high-performance computing (HPC) resources at OLCF, ALCF, and NERSC, as well as User sites such as PNNL and JLAB. We look at detecting and quantifying changes in site activity using a combination of t-Distributed Stochastic Neighbor Embedding (t-SNE) and decision tree analysis. Our findings bring insights into the working patterns and impact on data volume movements, particularly during late-night hours and weekends.

Collaboration in the IETF

RFC 9000, published in May 2021, marks an important milestone for the Internet's standardization body, the Internet Engineering Task Force (IETF): finally, the specification of the QUIC protocol is available. QUIC is the result of a five-year effort - and it is also the second of two major protocols (the first being SPDY, which later became HTTP/2) that Google LLC first deployed, and then brought to the IETF for standardization. This begs the question: when big players follow such a "shoot first, discuss later" approach, is IETF collaboration still "real", or is the IETF now being (mis-)used to approve protocols for standardization when they are already practically established, without really actively involving anyone but the main proponents?

The graph neural networking challenge

During the last decade, Machine Learning (ML) has increasingly become a hot topic in the field of Computer Networks and is expected to be gradually adopted for a plethora of control, monitoring and management tasks in real-world deployments. This poses the need to count on new generations of students, researchers and practitioners with a solid background in ML applied to networks. During 2020, the International Telecommunication Union (ITU) has organized the "ITU AI/ML in 5G challenge", an open global competition that has introduced to a broad audience some of the current main challenges in ML for networks. This large-scale initiative has gathered 23 different challenges proposed by network operators, equipment manufacturers and academia, and has attracted a total of 1300+ participants from 60+ countries. This paper narrates our experience organizing one of the proposed challenges: the "Graph Neural Networking Challenge 2020". We describe the problem presented to participants, the tools and resources provided, some organization aspects and participation statistics, an outline of the top-3 awarded solutions, and a summary with some lessons learned during all this journey. As a result, this challenge leaves a curated set of educational resources openly available to anyone interested in the topic.

A square law revisited

An earlier study observed that until 2008, the size of the BGP4 system for IPv4 appeared to have grown approximately in proportion to the square root of the host count of the globally addressable Internet. This article revisits this study by including IPv4 data until 2020 and adding IPv6 data. The results indicate that BGP4 for IPv4 is continuing to scale steadily even as IPv4 approaches its end of life, and that it is working as it should for IPv6, except for a slight concern that the number of announced routes is trending upwards faster as time goes on.

NemFi

This paper presents NemFi: a trace-driven WiFi emulator. NemFi is a record-and-replay emulator that captures traces representing real WiFi conditions, and later replay these traces to reproduce the same conditions. In this paper, we demonstrate that the state-of-the-art emulator that was developed for cellular links cannot emulate WiFi conditions. We identify the three key differences that must be addressed to enable accurate WiFi record-and-replay: WiFi packet losses, medium-access control, and frame aggregation. We then extend the existing cellular network emulator to support WiFi record-and-replay. We evaluate the performance of NemFi via repeated experimentation across different WiFi conditions and for three different types of applications: speed-test, file download, and video streaming. Our experimental results demonstrate that average application performance over NemFi and real WiFi links is similar (with less than 3 percent difference).