INFOCOM 2023 Papers

Many crowdsourcing platforms are emerging, leveraging the resources of recruited workers to execute various outsourcing tasks, mainly for those computing-intensive video analytics with high quality requirements. Although the profit of each platform is strongly related to the quality of analytics feedback, due to the uncertainty on diverse performance of workers and the conflicts of interest over platforms, it is non-trivial to determine the dispatch of tasks with maximum benefits. In this paper, we design a decentralized mechanism for a Crowd of Crowdsourcing platforms, denoted as Crowd2, optimizing the worker selection to maximize the social welfare of these platforms in a long-term scope, under the consideration of both proportional fairness and dynamic flexibility. Concretely, we propose a video analytics dispatch algorithm based on multi-agent bandit, for which the more accurate profit estimates are attained via the decoupling of multi-knapsack based mapping problem. Via rigorous proofs, a sub-linear regret bound for social welfare of crowdsourcing profits is achieved while both fairness and flexibility are ensured. Extensive trace-driven experiments demonstrate that Crowd2improves the social welfare by 36.8%, compared with other alternatives.

To achieve a line rate in the high-speed environment of modern networks, there is a continuing effort to offload network functions from software to programmable hardware (HW). Although the offloading effort has led to greater performance, it brings difficulty in the verification of timing-related network functions (Time-NFs) as well. Time-NFs use numerical timing values to perform various network tasks. For example, congestion control algorithm BBR uses round-trip time to improve throughput. Errors in Time-NFs could cause packet loss and poor throughput. However, verifying Time-NFs in HW often involves many clock cycles that can result in an exponentially increasing number of test cases. Current verification methods either do not scale or sacrifice soundness for scalability.In this paper, we propose an invariant-based method to improve the verification efficiency without losing soundness. Our method is motivated by an observation that most Time-NFs follow a few fixed patterns to use timing information. Based on these patterns, we develop a set of easy-to-validate invariants to constrain the examination space. According to experiments on real Time-NFs, our method can speed up verification by 7 times on average without losing the verification soundness.

With increasing concerns on privacy leakage from gradients, a variety of attack mechanisms emerged to recover private data from gradients at an honest-but-curious server, which challenged the primary advantage of privacy protection in federated learning. However, we cast doubt upon the real impact of these gradient attacks on production federated learning systems. By taking away several impractical assumptions that the literature has made, we find that gradient attacks pose a limited degree of threat to the privacy of raw data.Through a comprehensive evaluation on existing gradient attacks in a federated learning system with practical assumptions, we have systematically analyzed their effectiveness under a wide range of configurations. We present key priors required to make the attack possible or stronger, such as a narrow distribution of initial model weights, as well as inversion at early stages of training. We then propose a new lightweight defense mechanism that provides sufficient and self-adaptive protection against time-varying levels of the privacy leakage risk throughout the federated learning process. As a variation of gradient perturbation method, our proposed defense, called Outpost, selectively adds Gaussian noise to gradients at each update iteration according to the Fisher information matrix, where the level of noise is determined by the privacy leakage risk quantified by the spread of model weights at each layer. To limit the computation overhead and training performance degradation, Outpost only performs perturbation with iteration-based decay. Our experimental results demonstrate that Outpost can achieve a much better tradeoff than the state-of-the-art with respect to convergence performance, computational overhead, and protection against gradient attacks.

5G and beyond cellular networks (NextG) will support the continuous execution of resource-expensive edge-assisted deep learning (DL) tasks. To this end, Radio Access Network (RAN) resources will need to be carefully "sliced" to satisfy heterogeneous application requirements while minimizing RAN usage. Existing slicing frameworks treat each DL task as equal and inflexibly define the resources to assign to each task, which leads to sub-optimal performance. In this paper, we propose SEM-O-RAN, the first semantic and flexible slicing framework for NextG Open RANs. Our key intuition is that different DL classifiers can tolerate different levels of image compression, due to the semantic nature of the target classes. Therefore, compression can be semantically applied so that the networking load can be minimized. Moreover, flexibility allows SEM-O-RAN to consider multiple edge allocations leading to the same task-related performance, which significantly improves system-wide performance as more tasks can be allocated. First, we mathematically formulate the Semantic Flexible Edge Slicing Problem (SF-ESP), demonstrate that it is NP-hard, and provide an approximation algorithm to solve it efficiently. Then, we evaluate the performance of SEM-O-RAN through extensive numerical analysis with state-of-the-art multi-object detection (YOLOX) and image segmentation (BiSeNet V2), as well as real-world experiments on the Colosseum testbed. Our results show that SEM-O-RAN improves the number of allocated tasks by up to 169% with respect to the state of the art.

Existing solutions to network scheduling typically assume that the instantaneous link rates are completely known before a scheduling decision is made or consider a bandit setting where the accurate link quality is discovered only after it has been used for data transmission. In practice, the decision maker can obtain (relatively accurate) channel information, e.g., through beamforming in mmWave networks, right before data transmission. However, frequent beamforming incurs a formidable overhead in densely deployed mmWave WLANs. In this paper, we consider the important problem of throughput optimization with joint link probing and scheduling. The problem is challenging even when the link rate distributions are pre-known (the offline setting) due to the necessity of balancing the information gains from probing and the cost of reducing the data transmission opportunity. We develop an approximation algorithm with guaranteed performance when the probing decision is non-adaptive and a dynamic programming-based solution for the more challenging adaptive setting. We further extend our solutions to the online setting with unknown link rate distributions and develop a contextual-bandit based algorithm and derive its regret bound. Numerical results using data traces collected from real-world mmWave deployments demonstrate the efficiency of our solutions.

The ability to forecast mobile traffic patterns is key to resource management for mobile network operators and planning for local authorities. Several Deep Neural Networks (DNN) have been designed to capture the complex spatio-temporal characteristics of mobile traffic patterns at scale. These models are complex black boxes whose decisions are inherently hard to explain. Even worse, they have proven vulnerable to adversarial attacks which undermine their applicability in production networks. In this paper, we conduct a first in-depth study of the vulnerabilities of DNNs for large-scale mobile traffic forecasting. We propose DeExp, a new tool that leverages EXplainable Artificial Intelligence (XAI) to understand which Base Stations (BSs) are more influential for forecasting from a spatio-temporal perspective. This is challenging as existing XAI techniques are usually applied to computer vision or natural language processing and need to be adapted to the mobile network context. Upon identifying the more influential BSs, we run state-of-the art Adversarial Machine Learning (AML) techniques on those BSs and measure the accuracy degradation of the predictors. Extensive evaluations with real-world mobile traffic traces pinpoint that attacking BSs relevant to the predictor significantly degrades its accuracy across all the scenarios.

Quantum switches are envisioned to be an integral component of future entanglement distribution networks. They can provide high quality entanglement distribution service to end-users by performing quantum operations such as entanglement swapping and entanglement purification. In this work, we characterize the capacity region of such a quantum switch under noisy channel transmissions and imperfect quantum operations. We express the capacity region as a function of the channel and network parameters (link and entanglement swap success probability), entanglement purification yield and application level parameters (target fidelity threshold). In particular, we provide necessary conditions to verify if a set of request rates belong to the capacity region of the switch. We use these conditions to find the maximum achievable end-to-end user entanglement generation throughput by solving a set of linear optimization problems. We develop a max-weight scheduling policy and prove that the policy stabilizes the switch for all feasible request arrival rates. As we develop scheduling policies, we also generate new results for computing the conditional yield distribution of different classes of purification protocols. The conclusions obtained in this work can yield useful guidelines for subsequent quantum switch designs.

Kronecker-factored approximate curvature (K-FAC) has been shown to achieve faster convergence than SGD in training deep neural networks. However, existing distributed K-FAC (D-KFAC) relies on the all-reduce collective for communications and scheduling, which incurs excessive communications in each iteration. In this work, we propose a new form of D-KFAC with a reduce-based alternative to eliminate redundant communications. This poses new challenges and opportunities in that the reduce collective requires a root worker to collect the results, which considerably complicates the communication scheduling. To this end, we formulate an optimization problem that determines tensor fusion and tensor placement simultaneously aiming to minimize the training iteration time. We develop novel communication scheduling strategies and propose a placement-aware D-KFAC (PAD-KFAC) training algorithm, which further improves communication efficiency. Our experimental results on a 64-GPU cluster interconnected with 10Gb/s and 100Gb/s Ethernet show that our PAD-KFAC can achieve an average of 27% and 17% improvement over state-of-the-art D-KFAC methods, respectively.

Cloud network serves a large number of tenants and a variety of applications. The continuously 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, and lack of the capacity to realize complicated computation. The programmable switches can not satisfy the demand for 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 propose essential design and implementation challenges including a programming abstraction that allows automatically and effectively deploying 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 self-defined functions in a realistic system has verified the feasibility and practicality of the proposed framework.

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 transmitter or receiver individually. 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. Different from previous works, the proposed framework not only allows modeling all impairments of multiple devices, but also provides a new venue for joint compensation of the transmitter and receiver simultaneously. The solution has been successfully verified by the high baud rate (120Gbaud) coherent optical professional test platform and shows impressive optical Signalto-Noise Ratio (SNR) gains.

Gradient compression algorithms are widely used to alleviate the communication bottleneck in distributed ML. However, existing gradient compression algorithms suffer from accuracy degradation in Non-IID scenarios, because a uniform compression scheme is used to compress gradients at workers with different data distributions and volumes, since workers with larger volumes of data are forced to adapt to the same aggressive compression ratios as others. Assigning different compression ratios to workers with different data distributions and volumes is thus a promising solution. In this study, we first derive a function from capturing the correlation between the number of training iterations for a model to converge to the same accuracy, and the compression ratios at different workers; This function particularly shows that workers with larger data volumes should be assigned with higher compression ratios1to guarantee better accuracy. Then, we formulate the assignment of compression ratios to the workers as an n-variables chi-square nonlinear optimization problem under fixed and limited total communication constrain. We propose an adaptive gradient compression strategy called DAGC, which assigns each worker a different compression ratio according to their data volumes. Our experiments confirm that DAGC can achieve better performance facing highly imbalanced data volume distribution and restricted communication.

This paper discusses the problem of deploying a minimum number of directional sensors equipped with directional sensing units and directional communication antennas with beam-width ${\theta _c} \geq \frac{\pi }{2}$ such that the set of sensors covers a set of targets P in the 2D plane and the set of sensors forms a symmetric connected communication graph. As this problem is NP-hard, we propose an approximation algorithm that uses up to 3.5 times the number of omni-directional sensors required by the currently best approximation algorithm proposed by Han et al. [1]. This is a significant result since we have broken the barrier of $2\pi /\frac{\pi }{2} = 4$ when switching from omni-directional sensors to directional ones. Moreover, we improve the approximation ratio of the Strip-based algorithm for the Geometric Sector Cover problem proposed by Han et al. [1] from 9 to 7, and we believe that this result is of interest in the area of Computation Geometry. Simulation results show that our algorithms only require around 3 times the number of sensors used by Han et al.’s algorithm and significantly outperform other heuristics in practice.

Massive failures in communication networks are a consequence of natural disasters, heavy blackouts, military and cyber attacks. We tackle the problem of minimizing the time and number of interventions to sufficiently restore the communication network so as to support emergency services after large-scale failures. We propose PRoTOn (Progressive RecOvery and Tomography-based mONitoring), an efficient algorithm for progressive recovery of emergency services. Unlike previous work, assuming centralized routing and complete network observability, PRoTOn addresses the more realistic scenario in which the network relies on the existing routing protocols, and knowledge of the network state is partial and uncertain. Simulation results carried out on real topologies show that our algorithm outperforms previous solutions in terms of cumulative routed flow, repair costs and recovery time in both static and dynamic failure scenarios.

Software-defined networks are useful for multiple tasks, including firewalling, telemetry, and flow analysis. In particular, the P4 language makes it possible to carry out some simple packet processing tasks in the data plane, i.e., on the switch itself (without real-time support from the SDN controller or a server). However, owing to the limitations of packet parsing in P4, these tasks involve only the packet headers. In this paper, we present a novel approach that allows Deep Packet Inspection (DPI) – i.e., inspection of the packet payload – in the data plane, using P4 alone. We make use of the fact that in P4, a switch can clone and recirculate packets. One copy (clone) can be recirculated, slicing off a byte in each round, and using a finite-state machine to check if a target string has yet been seen. If the target string is found, the other copy (original packet) is discarded; if not, it is passed through. Our approach allows us to build the first application-layer firewall (URL filter) in the data plane, and to achieve essentially line-rate performance while filtering thousands of URLs, on a commodity programmable switch. It may in future also be used for other DPI tasks.

Distributed stream systems provide low latency by processing data as it arrives. However, existing systems do not provide latency guarantee, a critical requirement of real-time analytics, especially for stateful operators under burst and skewed workload. We present StreamSwitch, a control plane for stream systems to bound operator latency while optimizing resource usage. Based on a novel stream switch abstraction that unifies dynamic scaling and load balancing into a holistic control framework, our design incorporates reactive and predictive metrics to deduce the healthiness of executors and prescribes practically optimal scaling and load balancing decisions in time. We implement a prototype of StreamSwitch and integrate it with Apache Flink and Samza. Experimental evaluations on real-world applications and benchmarks show that StreamSwitch provides cost-effective solutions for bounding latency and outperforms the state-of-the-art alternative solutions.