INFOCOM 2023 Papers

RDMA provides extremely low-latency and high- throughput data transmission as its protocol stack is entirely offloaded into the RDMA NIC. However, the increasing scale of RDMA networks requires hosts to establish a large number of connections, e.g., process-level full mesh, which easily overwhelms the limited resource on RNICs and hence significantly degrades performance. This paper presents SRM, a scalable transport mode for RDMA that remarkably alleviates resource exhaustion on RNICs. SRM proposes a kernel-based solution to multiplex workloads from different applications over the same connection. Meanwhile, to preserve RDMA’s performance benefits, SRM 1) avoids syscall overhead by sharing the working memory between user-space and kernel; 2) maintains high resource utilization through lock-free approach to avoid contention; 3) adopts multiple optimizations to mitigate the head-of-line blocking issue; 4) implements a rapid recovery mechanism to provide high system robustness. We evaluate SRM using extensive experiments and simulations. Testbed experiments reveal that SRM outperforms existing transports, including DCT, RC, and XRC, by 4x to 20x in latency for all-to-all communication pattern. Simulations of large-scale networks show that, compared with DCT, RC, and XRC, SRM achieves up to 4.42x/4.0x/3.7x speedups respectively in flow completion time while consuming the least memory.

This paper presents our design and implementation of a fast and accurate data aggregation strategy for LoRa networks named One-shot. To facilitate data aggregation, One-shot assigns distinctive chirps for different LoRa nodes to encode individual data. One-shot coordinates the nodes to concurrently transmit encoded packets. Receiving concurrent transmissions, One-shot gateway examines the frequencies of superimposed chirp signals and computes application-defined aggregate functions (e.g., sum, max, count, etc.), which give a quick overview of sensor data in a large monitoring area. One-shot develops techniques to handle a series of practical challenges involved in frequency and time synchronization of concurrent chirps. We evaluate the effectiveness of One-shot with extensive experiments. Results show that One-shot substantially outperforms state-of-the-art data aggregation methods in terms of aggregation accuracy as well as query efficiency.

By combing edge computing and parallel computing, distributed edge computing has emerged as a new paradigm to accelerate computation at the edge. Considering the parallelism of both computation and communication, the problem of Minimum Latency joint Communication and Computation Scheduling (MLCCS) is studied recently. However, existing works have rigid assumptions that the communication time of each device is fixed and the workload can be split arbitrarily small. Aiming at making the work more practical and general, the MLCCS problem without the above assumptions is studied in this paper. Firstly, the MLCCS problem under a general model is formulated and proved to be NP-hard. Secondly, a pyramid-based computing model is proposed to consider the parallelism of communication and computation jointly, which has an approximation ratio of 1 + δ, where δ is related to devices’ communication rates. An interesting property under such computing model is identified and proved, i.e., the optimal latency can be obtained under arbitrary scheduling order when all the devices share the same communication rate. When the devices own different communication rates, the optimal scheduling order is also obtained. Additionally, when the workload cannot be split arbitrarily, an approximation algorithm with ratio of at most 2 (1 + δ) is proposed. Finally, the theoretical analysis and simulation results verify that the proposed algorithm has high performance in terms of latency. Two testbed experiments are also conducted, which show that the proposed method outperforms the existing methods, reducing the latency by up to 29.2% in real-world applications.

How to allocate network paths and their resources to minimize the delivery time of data transfer tasks over time-varying networks? Solving this MDDT (Minimizing Data Delivery Time) problem has important applications from data centers to delay-tolerant networking. In particular, with the rapid deployment of satellite networks in recent years, an efficient MDDT solver will serve as a key building block there.The MDDT problem can be solved in polynomial time by finding the maximum flow in a time-expanded graph. A binary-search-based solver incurs O(N•log N•Γ) time complexity, where N corresponds to time horizon and Γ is the time complexity to solve a maximum flow problem for one snapshot of the network. In this work, we design a one-pass solver that progressively expands the graph over time until it reaches the earliest time interval n to complete the delivery. By reusing the calculated maximum flow results from earlier iterations, it solves the MDDT problem while incurring only O(nΓ) time complexity for algorithms that can apply our technique. We apply the one-pass design to Ford-Fulkerson algorithm and evaluate our solver using a network of 184 satellites from Starlink constellations. We demonstrate >75× speed-up in the running time and show that our solution can also enable advanced applications such as preemptive scheduling.

Federated learning protects users’ data privacy though sharing users’ local model parameters (instead of raw data) with a server. However, when massive users train a large machine learning model through federated learning, the dynamically varying and often heavy communication overhead can put significant pressure on the network operator. The operator may choose to dynamically change the network prices in response, which will eventually affect the payoffs of the server and users. This paper considers the under-explored yet important issue of the joint design of participation incentives (for encouraging users’ contribution to federated learning) and network pricing (for managing network resources). Due to heterogeneous users’ private information and multi-dimensional decisions, the optimization problems in Stage I of multi-stage games are non-convex. Nevertheless, we are able to analytically derive the corresponding optimal contract and pricing mechanism through proper transformations of constraints, variables, and functions, under both vertical and horizontal interaction structures of the participants. We show that the vertical structure is better than the horizontal one, as it avoids the interests misalignment between the server and the network operator. Numerical results based on real-world datasets show that our proposed mechanisms decrease server’s cost by up to 24.87% comparing with the state-of-the-art benchmarks.

For-hire vehicle-enabled crowd sensing (FVCS) has become a promising paradigm to conduct urban sensing tasks in recent years. FVCS platforms aim to jointly optimize both the order-serving revenue as well as sensing coverage and quality. However, such two objectives are often conflicting and need to be balanced according to the platforms’ preferences on both objectives. To address this problem, we propose a novel cooperative multi-objective multi-agent reinforcement learning framework, referred to as MOVDN, to serve as the first preference-configurable order dispatch mechanism for FVCS platforms. Specifically, MOVDN adopts a decomposed network structure, which enables agents to make distributed order selection decisions, and meanwhile aligns each agent’s local decision with the global objectives of the FVCS platform. Then, we propose a novel algorithm to train a single universal MOVDN that is optimized over the space of all preferences. This allows our trained model to produce the optimal policy for any preference. Furthermore, we provide the theoretical convergence guarantee and sample efficiency analysis of our algorithm. Extensive experiments on three real-world ride-hailing order datasets demonstrate that MOVDN outperforms strong baselines and can support the platform in decision-making effectively.

Network operation and maintenance rely heavily on network traffic monitoring. Due to the measurement overhead reduction, lack of measurement infrastructure, and unexpected transmission error, network traffic monitoring systems suffer from incomplete observed data and high data sparsity problems. Recent studies model missing data recovery as a tensor completion task and show good performance. Although promising, the current tensor completion models adopted in network traffic data recovery lack an effective and efficient retraining scheme to adapt to newly arrived data while retaining historical information. To solve the problem, we propose LightNestle, a novel sequential tensor completion scheme based on meta-learning, which designs (1) an expressive neural network to transfer spatial knowledge from previous embeddings to current embeddings; (2) an attention-based module to transfer temporal patterns into current embeddings in linear complexity; and (3) meta-learning-based algorithms to iteratively recover missing data and update transfer modules to catch up with learned knowledge. We conduct extensive experiments on two real-world network traffic datasets to assess our performance. Results show that our proposed methods achieve both fast retraining and high recovery accuracy.

Provenance tracking has been widely used in the recent literature to debug system vulnerabilities and find the root causes behind faults, errors, or crashes over a running system. However, the existing approaches primarily developed graph-based models for provenance tracking over monolithic applications running directly over the operating system kernel. In contrast, the modern DevOps-based service-oriented architecture relies on distributed platforms, like serverless computing that uses container-based sandboxing over the kernel. Provenance tracking over such a distributed micro-service architecture is challenging, as the application and system logs are generated asynchronously and follow heterogeneous nomenclature and logging formats. This paper develops a novel approach to combining system and micro-services logs together to generate a Universal Provenance Graph (UPG) that can be used for provenance tracking over serverless architecture. We develop a Loadable Kernel Module (LKM) for runtime unit identification over the logs by intercepting the system calls with the help from the control flow graphs over the static application binaries. Finally, we design a regular expression-based log optimization method for reverse query parsing over the generated UPG. A thorough evaluation of the proposed UPG model with different benchmarked serverless applications shows the system’s effectiveness.

Federated learning (FL) enables geographically dispersed edge devices (i.e., clients) to learn a global model without sharing the local datasets, where each client performs gradient descent with its local data and uploads the gradients to a central server to update the global model. However, FL faces massive communication overhead resulted from uploading the gradients in each training round. To address this problem, most existing research compresses the gradients with fixed and unified quantization for all the clients, which neither seeks adaptive quantization due to the varying gradient norms at different rounds, nor exploits the heterogeneity of the clients to accelerate FL. In this paper, we propose a novel adaptive and heterogeneous gradient quantization algorithm (AdaGQ) for FL to minimize the wall-clock training time from two aspects: i) adaptive quantization which exploits the change of gradient norm to adjust the quantization resolution in each training round; and ii) heterogeneous quantization which assigns lower quantization resolution to slow clients to align their training time with other clients to mitigate the communication bottleneck, and higher quantization resolution to fast clients to achieve a better communication efficiency and accuracy tradeoff. Evaluations based on various models and datasets validate the benefits of AdaGQ, reducing the total training time by up to 52.1% compared to baseline algorithms (e.g., FedAvg, QSGD).

The RF environment in a secure space can be compromised by intentional transmissions of hard-to-detect underlay signals that overlap with a high-power baseline transmission. Specifically, we consider the case where a direct sequence spread spectrum (DSSS) signal is the underlay signal hiding within a baseline 4G Long-Term Evolution (LTE) signal. As compared to overt actions like jamming, the DSSS signal allows the LTE signal to be decodable, which makes it hard to detect. ICARUS presents a machine learning based framework that offers choices at the physical layer for inference with inputs of (i) in-phase and quadrature (IQ) samples only, (ii) cycle-frequency features obtained via cyclostationary signal processing (CSP), and (iii) fusion of both, to detect the underlay DSSS signal and its modulation type within LTE frames. ICARUS chooses the best inference method considering both the expected accuracy and the computational overhead. ICARUS is rigorously validated on multiple real-world datasets that include signals captured in cellular bands in the wild and the NSF POWDER testbed for advanced wireless research (PAWR). Results reveal that ICARUS can detect DSSS anomalies and its modulation scheme with 98-100% and 67 − 99% accuracy, respectively, while completing inference within 3 − 40 milliseconds on an NVIDIA A100 GPU platform.

A variety of network management and orchestration (MANO) tasks take advantage of predictions to support anticipatory decisions. In many practical scenarios, such predictions entail two largely overlooked challenges: (i) the exact relationship between the predicted values (e.g., reserved resources) and the performance objective (e.g., quality of experience of end users) is often tangled and cannot be known a priori, and (ii) the objective is linked in many cases to multiple predictions that contribute to it in an intertwined way (e.g., resources to reserved are limited and must be shared among competing flows). We present AutoManager, a novel meta-learning model that can support complex MANO tasks by addressing these two challenges. Our solution learns how multiple intertwined predictions affect a common performance goal, and steers them so as to attain the correct operation point under a-priori unknown loss functions. We demonstrate AutoManager in practical, complex use cases based on real-world traffic measurements; our experiments show that the model produces forecasts that are accurate and tailored to the MANO task in a fully automated way.

Spectrum crowdsensing systems do not provide labeled data near real-time yet. We propose a framework that addresses this challenge and relies solely on Power Spectrum Density (PSD) data collected by low-cost receivers. A major hurdle is to design a system that is computationally efficient for near real-time operation, yet using only the limited 2 MHz bandwidth of low-cost spectrum sensors. First, we present a method for unsupervised transmission detection that works with PSD data already collected by the backend of the crowdsensing platform, and that provides stable detection of transmission boundaries. Second, we introduce a data-driven deep learning solution to classify the wireless technology used by the transmitter, using transmission features in a compressed space extracted from single PSD measurements over at most 2 MHz band. We build an experimental platform, and evaluate our framework with real-world data collected from 47 different sensors deployed across Europe. We show that our framework yields an average classification accuracy close to 94.25% over the testing dataset, with a maximum latency of 3.4 seconds when integrated in the backend of a major crowdsensing network. Code and data have been released for reproducibility and further studies.

Virtual reality (VR) can provide users with an immersive experience in the metaverse. One of the most promising properties of VR is that users’ identities can be protected by changing their physical world appearances into arbitrary virtual avatars. However, recent proposed de-anonymization attacks demonstrate the feasibility of recognizing the user’s identity behind the VR avatar’s masking. In this paper, we propose AvatarHunter, a non-intrusive and user-unconscious de-anonymization attack based on victims’ inherent movement signatures. AvatarHunter imperceptibly collects the victim avatar’s gait information via recording videos from multiple views in the VR scenario without requiring any permission. A Unity-based feature extractor is designed that preserves the avatar’s movement signature while immune to the avatar’s appearance changes. Real-world experiments are conducted in VRChat, one of the most popular VR applications. The experimental results demonstrate that AvatarHunter can achieve attack success rates of 92.1% and 66.9% in closed-world and open-world avatar settings, respectively, which are much better than existing works.

IEEE INFOCOM 2023 - IEEE Conference on Computer Communications, New York City, NY, USA, 17-20 May 2023