Session C-4


8:30 AM — 10:00 AM EDT
May 18 Thu, 8:30 AM — 10:00 AM EDT
Babbio 202

SkyNet: Multi-Drone Cooperation for Real-Time Identification and Localization

Junkun Peng (Tsinghua University, China); Qing Li (Peng Cheng Laboratory, China); Yuanzheng Tan (Sun Yat-Sen University, China); Dan Zhao (Peng Cheng Laboratory, China); Zhenhui Yuan (Northumbria University, United Kingdom (Great Britain)); Jinhua Chen (Sun Yat-Sen University, China); Hanling Wang (Tsinghua University & Peng Cheng Laboratory, China); Yong Jiang (Graduate School at Shenzhen, Tsinghua University, China)

Aerial images from drones have been used to detect and track suspects in the crowd for the public safety purpose. However, using a single drone for human identification and localization faces many challenges including low accuracy and long latency, due to poor visibility, varying field of views (FoVs) and limited on-board computing resources. In this paper, we propose SkyNet, a multi-drone cooperative system for accurate and realtime human identification and localization. SkyNet computes the 3D position of a person by cross searching from multiple views. To achieve high accuracy in identification, SkyNet fuses aerial images of multiple drones according to their legibility. Moreover, by predicting the estimated finishing time of tasks, SkyNet schedules and balances workloads among edge devices and the cloud server to minimize processing latency. We implement and deploy SkyNet in real life, and evaluate the performance of identification and localization with 20 human participants. The results show that SkyNet can locate people with an average error within 0.18m on a square of 554m2 . The identification accuracy reaches 91.36%, and the localization and identification process is completed within 0.84s.
Speaker Junkun Peng (Tsinghua University)

Junkun Peng received his B.S. degree in Information Management and Information Systems from Shanghai University, China, in 2015.

He is currently pursuing a Ph.D. in Computer Science at Tsinghua University, China.

His research focuses on real-time video transmission and analysis, and robot learning.

A2-UAV: Application-Aware Content and Network Optimization of Edge-Assisted UAV Systems

Andrea Coletta (JP Morgan AI Research, Italy); Flavio Giorgi, Gaia Maselli, Matteo Prata and Domenicomichele Silvestri (Sapienza University of Rome, Italy); Jonathan Ashdown (United States Air Force, USA); Francesco Restuccia (Northeastern University, USA)

To perform object/person detection and target tracking, Unmanned Autonomous Vehicles (UAVs) require the continuous execution of edge-assisted vision tasks based on Deep Neural Networks (DNNs). In multi-hop UAV networks, the execution of these tasks fall short of expectations, owing to the typical bandwidth constraints which affect the quality of received data. In this paper, we propose a novel framework to optimize the number of correctly executed tasks. We consider the UAV application at hand (i.e., the specific DNN and the classes of interest) and formulate an optimization problem that takes into account (i) the relationship between DNN accuracy and image resolution for the classes of interest based on the available dataset, (ii) the target positions, (iii) the current energy/position of the UAVs to optimize routing, data pre-processing and target assignment for each UAV. We extensively evaluate our approach through large-scale simulations, as well as real-world experiments with a testbed. We consider state-of-the-art image classification tasks with four different DNN models. Results show that our approach outperform state-of-the-art approaches. To allow full reproducibility, we pledge to share datasets and code with the research community.
Speaker Matteo Prata (Sapienza University of Rome, Italy)

Matteo Prata is a PhD student in the Department of Computer Science at Sapienza University of Rome, Italy. He has authored numerous research papers in the field of computer network performance, with a particular focus on unmanned aerial vehicle networks. Additionally, his research interests encompass AI applications in finance.

WiSwarm: Age-of-Information-based Wireless Networking for Collaborative Teams of UAVs

Vishrant Tripathi (MIT, USA); Igor Kadota (Columbia University, USA); Ezra Tal (MIT, USA); Muhammad Shahir Abdurrahman (Stanford University & Massachusetts Institute of Technology, USA); Alexander Warren, Sertac Karaman and Eytan Modiano (MIT, USA)

The Age-of-Information (AoI) metric has been widely studied in the theoretical communication networks and queuing systems literature. However, experimental evaluation of its applicability to complex real-world time-sensitive systems is largely lacking. In this work, we develop, implement, and evaluate an AoI-based application layer middleware that enables the customization of WiFi networks to the needs of time-sensitive applications. By controlling the storage and flow of information in the underlying WiFi network, our middleware can: (i) prevent packet collisions; (ii) discard stale packets that are no longer useful; and (iii) dynamically prioritize the transmission of the most relevant information. To demonstrate the benefits of our middleware, we implement a mobility tracking application using a swarm of UAVs communicating with a central controller via WiFi. Our experimental results show that, when compared to WiFi-UDP/WiFi-TCP, the middleware can improve information freshness by a factor of 109x/48x and tracking accuracy by a factor of 4x/6x, respectively. Most importantly, our results also show that the performance gains of our approach increase as the system scales and/or the traffic load increases.
Speaker Vishrant Tripathi (MIT)

Vishrant Tripathi is a Ph.D. candidate in the EECS department at MIT, working with Prof. Eytan Modiano at the Laboratory for Information and Decision Systems (LIDS). His research is on modeling, analysis and design of communication networks, with emphasis on wireless and real-time networks. His current focus is on scheduling problems in networked control systems, multi-agent robotics and federated learning.

FlyTracker: Motion Tracking and Obstacle Detection for Drones Using Event Cameras

Yue Wu, Jingao Xu and Danyang Li (Tsinghua University, China); Yadong Xie (Beijing Institute of Technology, China); Hao Cao (Tsinghua University, China); Fan Li (Beijing Institute of Technology, China); Zheng Yang (Tsinghua University, China)

Location awareness in environments is one of the key parts for drones' applications and have been explored through various visual sensors. However, standard cameras easily suffer from motion blur under high moving speeds and low-quality image under poor illumination, which brings challenges for drones to perform motion tracking. Recently, a kind of bio-inspired sensors called event cameras emerge, offering advantages like high temporal resolution, high dynamic range and low latency, which motivate us to explore their potential to perform motion tracking in limited scenarios. In this paper, we propose FlyTracker, aiming at developing visual sensing ability for drones of both individual and circumambient location-relevant contextual, by using a monocular event camera. In FlyTracker, background-subtraction-based method is proposed to distinguish moving objects from background and fusion-based photometric features are carefully designed to obtain motion information. Through multilevel fusion of events and images, which are heterogeneous visual data, FlyTracker can effectively and reliably track the 6-DoF pose of the drone as well as monitor relative positions of moving obstacles. We evaluate performance of FlyTracker in different environments and the results show that FlyTracker is more accurate than the state-of-the-art baselines.
Speaker Yue Wu

Yue Wu received the doctor degree in Computer Science and Technology from Beijing Institute of Technology, in 2021. Currently she is a post-doctoral in the School of Software, Tsinghua University, Beijing, China. Her research interests include visual localization, mobile computing, and Internet of things.

Session Chair

Enrico Natalizio

Session C-5

Optical and Mobile

1:30 PM — 3:00 PM EDT
May 18 Thu, 1:30 PM — 3:00 PM EDT
Babbio 202

AGO: Boost Mobile AI Inference Performance by Removing Constraints on Graph Optimization

Zhiying Xu, Hongding Peng and Wei Wang (Nanjing University, China)

Traditional deep learning compilers rely on heuristics for subgraph generation, which impose extra constraints on graph optimization, e.g., each subgraph can only contain at most one complex operator. In this paper, we propose AGO, a framework for graph optimization with arbitrary structures to boost the inference performance of deep models by removing such constraints. To create new optimization opportunities for complicated subgraphs, we propose intensive operator fusion, which can effectively stitch multiple complex operators together for better performance. Further, we design a graph partitioning scheme that allows an arbitrary structure for each subgraph while guaranteeing the acyclic property among all generated subgraphs. Additionally, to enable efficient performance tuning on complicated subgraphs, we devise a novel divide-and-conquer tuning mechanism to orchestrate different system components. Through extensive experiments on various neural networks and mobile devices, we show that our system can improve the inference performance by up to 3.3× when compared with state-of-the-art deep compilers.
Speaker Zhiying Xu (Nanjing University)

I am a PhD candidate at Nanjing University, China. I am interested in machine learning system, compilation, and code generation techniques.

OpticNet: Self-Adjusting Networks for ToR-Matching-ToR Optical Switching Architectures

Caio Alves Caldeira (Universidade Federal de Minas Gerais, Brazil); Otavio Augusto de Oliveira Souza and Olga Goussevskaia (UFMG, Brazil); Stefan Schmid (University of Vienna, Austria)

Demand-aware reconfigurable datacenter networks can be modeled as a ToR-Matching-ToR (TMT) two-layer architecture, in which each top-of-rack (ToR) is represented by a static switch, and $n$ ToRs are connected by a set of reconfigurable optical circuit switches (OCS). Each OCS internally connects a set of in-out ports via a matching that may be updated at runtime. The matching model is a formalization of such networks, where the datacenter topology is defined by the union of matchings over the set of nodes, each of which can be reconfigured at unit cost.
In this work we propose a scalable matching model for scenarios where OCS have a constant number of ports. Furthermore, we present OpticNet, a framework that maps a set of $n$ static ToR switches to a set of $p$-port OCS to form any constant-degree topology. We prove that OpticNet uses a minimal number of reconfigurable switches to realize any desired network topology and allows to apply any existing self-adjusting network (SAN) algorithm on top of it, also preserving amortized performance guarantees. Our experimental results based on real workloads show that OpticNet is a flexible and efficient framework to design efficient SANs.
Speaker Otávio Augusto de Oliveira Souza (Universidade Federal of Minas Gerais)

Otávio A. de O. Souza received the M.Sc degree in Computer Science in 2020 from Universidade Federal de Minas Gerais (UFMG), Brazil, where he is currently a Ph.D. candidate in Computer Science. His research interest includes modeling, algorithm design, and analysis in communication networks, with emphasis on distributed systems.

An End-to-end Learning Framework for Joint Compensation of Impairments in Coherent Optical Communication Systems

Rui Zhang (University of Electronic Science and Technology of China, China); Min Liao and Jun Chen (Huawei Chengdu Research Center, China); Xusong Ning, Lin Li and Qinli Yang (University of Electronic Science and Technology of China, China); Yongsheng Xu (Huawei Chengdu Research Center, China); Junming Shao (University of Electronic Science and Technology of China, China)

The application of machine learning techniques in Coherent Optical Communication (COC) systems has gained increasing attention in recent years. One representative and successful application is to employ neural networks to compensate the signal impairments of devices in the COC system. However, existing studies usually concentrate on each individual device or one impairment, the various impairments sourced from multiple devices (e.g., non-linear distortion, memory, and crosstalk effects) are not well investigated. More importantly, due to the impairment isolation caused by frequency offset, traditional studies only compensate the impairments of the transmitter or receiver individually. Different from previous works, in this paper, we consider a more practical and challenging experimental setup environment: joint compensation of multiple impairments associated with all devices of transmitter and receiver simultaneously. To this end, we propose an end-to-end compensation framework from the transmitter to the receiver in COC systems with three associated modules: an auxiliary channel neural network for impairment modeling, a pre-compensation neural network deployed in the transmitter, and a post-compensation neural network deployed in the receiver. The solution has been successfully verified by the high baud rate (120Gbaud) coherent optical professional test platform and shows impressive optical Signal-to-Noise Ratio (SNR) gains.
Speaker Rui Zhang( University of Electronic Science and Technology of China)

I am a Ph.D. candidate at the School of Computer Science of the USETC under the supervision of Prof. Junimg Shao. My main research area is adversarial machine learning and its application with a particular focus on adversarial defense and the deep learning model's robustness.

Minimizing Age of Information for Underwater Optical Wireless Sensor Networks

Yu Tian, Lei Wang, Chi Lin, Yang Chi, Bingxian Lu and Zhenquan Qin (Dalian University of Technology, China)

In Underwater Optical Wireless Sensor Networks (UOWSNs), the Age of information (AoI) is critical as it manifests information freshness, which is extremely useful for real-time monitoring applications. Prior works extensively focus on improving AoI in single-hop or multi-hop networks with fixed paths. However, the issue of multi-path transmission is overlooked. In this paper, we make the first attempt to address the issue of minimizing AoI through link scheduling in UOWSNs (termed MAKE problem). To minimize AoI, we propose an AoI model and corresponding link constraint model for multi-path and multi-hop UOWSNs. Then, we formalize the MAKE problem into a submodular function maximization problem, and propose a greedy method with an approximation ratio guarantee to solve it. Theoretical analyses are presented to explore the features of our scheme. Finally, to demonstrate the effectiveness of the proposed scheme, extensive simulations are conducted. The results reveal that the proposed scheme has 21.4% lower total average AoI compared with other baseline algorithms. Furthermore, test-bed experiments are carried out to verify the applicability of the proposed scheme in practical applications.
Speaker Yu Tian (Dalian University of Technology)

Yu Tian received an M.S. degree from Inner Mongolia University, Hohhot, China, in 2019. He is currently pursuing a Ph.D. degree in software engineering from the Dalian University of Technology, Dalian, China. His research interest focuses on the underwater optical wireless network.

Session Chair

Yan Zhang

Session C-6

Programmable Switches

3:30 PM — 5:00 PM EDT
May 18 Thu, 3:30 PM — 5:00 PM EDT
Babbio 202

Flowrest: Practical Flow-Level Inference in Programmable Switches with Random Forests

Aristide Tanyi-Jong Akem (IMDEA Networks Institute, Spain & Universidad Carlos III de Madrid, Spain); Michele Gucciardo and Marco Fiore (IMDEA Networks Institute, Spain)

User-plane machine learning enables low-latency and high-throughput inference at line rate. Yet, data planes are highly constrained environments, and restrictions are especially marked in programmable switches with limited available memory and minimum support for mathematical operations or data types. As a result, current solutions for in-switch inference that are compatible with production-level hardware lack support for complex features or suffer from limited scalability, which creates performance barriers in complex tasks involving large decision spaces. We address this limitation, and present Flowrest, a first complete Random Forest (RF) model implementation that can operate at the level of individual flows in commercial switches. Our solution builds on (i) a novel framework to embed generic flow-level machine learning models into programmable switch ASICs, and (ii) original guidelines for tailoring RF models to operations in programmable switches already at the design stage. We implement Flowrest as open-source software using the P4 language, and assess its performance in an experimental platform based on Intel Tofino switches. Tests with tasks of unprecedented complexity show how our model can improve accuracy by up to 39% over previous approaches to implement RF models in real- world equipment.
Speaker Aristide Tanyi-Jong Akem (IMDEA Networks Institute & Universidad Carlos III de Madrid)

Akem is a PhD student in the Networks Data Science Group at IMDEA Networks Institute in Madrid, Spain. He is also a student at Universidad Carlos III de Madrid, where he is enrolled in the Telematics Engineering program. Prior to his PhD studies, he completed an engineering degree at the University of Yaounde I, in Cameroon and a master's in electrical and computer engineering at Carnegie Mellon University Africa, in Rwanda. He is currently involved with the European Union's Horizon 2020 project "BANYAN" which aims to bring big data analytics to radio access networks. At the moment, he is visiting Orange Labs in Paris, France as part of the secondments of the project. Akem's current research interest is in the area of in-band network intelligence, with a focus on in-network machine learning.

Melody: Toward Resource-Efficient Packet Header Vector Encoding on Programmable Switches

Xiang Chen and Hongyan Liu (Zhejiang University, China); Qingjiang Xiao and Jianshan Zhang (Fuzhou University, China); Qun Huang (Peking University, China); Dong Zhang (Fuzhou University, China); Xuan Liu (Yangzhou University & Southeast University, China); Chunming Wu (College of Computer Science, Zhejiang University, China)

Emerging programmable switches provision a limited capacity of packet header vector (PHV) resources to support the user-specified logic of manipulating packet header fields and metadata fields in network functions. However, existing switch compilers typically employ inefficient strategies of encoding fields onto PHV words. These strategies significantly waste scarce PHV resources, leading to possible failures when offloading network functions. In this paper, we propose Melody, a framework that aims to achieve resource efficiency in PHV encoding. The key idea of Melody is to reuse PHV words for as many fields as possible to improve resource efficiency. To realize this idea, we design both a field analyzer and an optimization framework in Melody. The field analyzer automatically identifies which fields can reuse PHV words while preserving the original packet processing logic, avoiding laborious user burdens. The optimization framework integrates analysis results into PHV encoding to produce the resource-optimal decisions. We have implemented Melody and evaluated it with real network functions and large-scale synthetic ones. Our results indicate that compared to existing compilers, Melody improves the resource efficiency by up to 85%.
Speaker Xiang Chen

Xiang is a first-year PhD student at Zhejiang University. His advisors are Prof. Chunming Wu, Prof. Qun Huang, and Prof. Dong Zhang. He has received a Best Paper Award from IEEE/ACM IWQoS 2021 and a Best Paper Candidate from IEEE INFOCOM 2021. His research interests include programmable networks, network virtualization, and network security.

CLIP: Accelerating Features Deployment for Programmable Switch

Tingting Xu, Xiaoliang Wang and Chen Tian (Nanjing University, China); Yun Xiong and Yun Lin (HUAWEI, China); Baoliu Ye (Nanjing University, China)

Cloud network serves a large amount of tenants and a variety of applications. The continuous changing demands require a programmable data plane to achieve fast feature velocity. However, the years-long release cycle of traditional function-fixed switches can not meet this requirement. Emerging programmable switches provide the flexibility of packet processing without sacrificing hardware performance. Due to the trade-off between performance and flexibility, the current programmable switches make compromises in some aspects such as limited memory/computation resources, lack of the capacity to realize complicated computation. The programmable switches can not satisfy the demand of network services and applications in production networks. We propose a framework that leverages host servers to extend the capability of network switches quickly, accelerates new feature deployment, and verifies new ideas in production networks. Specifically, to build the unified programmable data plane, we address essential design and implementation challenges including a programming abstraction that allows automatically and effectively deploy network functions on switch and server clusters, allocating traffic to fully utilize the server resources, and supporting flexible scaling of the system. The quick deployment of several self-defined functions in realistic system have verified the feasibility and practicality of the proposed framework.
Speaker Tingting Xu (Nanjing University)

Tingting Xu received the B.E. degree in 2019 from the college of Computer Science and Electronic Engineering, Hunan University, Hunan, China. She is currently working toward the Ph.D. degree in the Department of Computer Science and Technology, Nanjing University under the supervision of Prof.Xiaoliang Wang. Her research interests include programmable network, datacenter network, and network function virtulization.

RED: Distributed Program Deployment for Resource-aware Programmable Switches

Xingxin Jia, Fuliang Li and Songlin Chen (Northeastern University, China); Chengxi Gao (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, China); Pengfei Wang (Dalian University of Technology, China); Xingwei Wang (Northeastern University, China)

Programmable switches allow data plane to program how packets are processed, which enables flexibility for network management tasks, e.g., packet scheduling and flow measurement. Existing studies focus on program deployment at a single switch, while deployment across the whole data plane is still a challenging issue. In this paper, we present RED, a Resource-Efficient and Distributed program deployment solution for programmable switches. First of all, we compile the data plane programs to estimate the resource utilization and divide them into two categories that wait to be further processed. Then, the proposed merging and splitting algorithms are selectively applied to merge or split the pending programs. Finally, we consolidate the scarce resources of the whole data plane to deploy the programs. Extensive experiment results show that 1) RED improves the speedup by two orders of magnitude compared to P4Visor and merges 58.64% more nodes than SPEED; 2) RED makes the overwhelmed programs run normally at a single switch and meanwhile reduces 3% latency of inter-device scheduling; 3) RED achieves network-wide resource balancing in a distributed way.
Speaker Xingxin Jia(Northeastern University, China)

Session Chair

Patrick P. C. Lee

Gold Sponsor

Gold Sponsor

Bronze Sponsor

Student Travel Grants

Student Travel Grants

Local Organizer

Made with in Toronto · Privacy Policy · INFOCOM 2020 · INFOCOM 2021 · INFOCOM 2022 · © 2023 Duetone Corp.