Checkpointing is becoming a hotspot of interest in both academia and industry as the primary fault-tolerance method for large model training. However, existing checkpoint designs are tightly coupled with the training process, leading to interruptions that reduce overall training efficiency. To reduce the impact of checkpoints on training, this paper presents FlowCheck, a novel checkpointing system that decouples checkpoint operations from the training process, enabling checkpoint saving without blocking the training. Specifically, FlowCheck updates the checkpoints by extracting complete gradient information from the network traffic of normal training. FlowCheck deploys a traffic-mirroring network to support this design. To utilize mirrored traffic for checkpointing operations, two key challenges need to be addressed. First, we need to achieve precise identification and extraction of gradient packets from training traffic. Second, the transmission on the mirror link is unreliable due to its inability to trigger retransmission upon packet loss. Through two key designs: (1) packet-counting-based traffic identification, and (2) packet redundancy recovery mechanism, FlowCheck implements an efficient checkpointing system using the existing training network and solves the above two challenges. Experiments and estimations verify that FlowCheck achieves checkpoint operations with zero impact on training, and demonstrate that FlowCheck achieves over 98% effective training time under practical fault conditions.

Single Root I/O Virtualization (SR-IOV) technology has advanced in recent years and can simultaneously satisfy the network requirements of high data plane performance, high deployment density, and fast startup for applications in traditional containers. However, it falls short with secure containers, which have become the mainstream choice in multi-tenant clouds. SR-IOV requires secure containers to use passthrough I/O for higher data plane performance, which hinders the container startup performance and prevents its usage in time-sensitive tasks like serverless computing. In this paper, we advocate that the startup performance of SR-IOV enabled secure containers can be further boosted, making SR-IOV suitable for building a Container Network Interface (CNI) for secure containers. We first dissect the end-to-end concurrent startup process and identify three key bottlenecks that lead to the slow startup, including Virtual Function I/O device set management, Direct Memory Access memory mapping, and Virtual Function (VF) driver initialization. We then propose a CNI named FastIOV that addresses these bottlenecks through lock decomposition, unnecessary mapping skipping, decoupled zeroing, and asynchronous VF driver initialization. Our evaluation shows that FastIOV reduces the overhead of enabling SR-IOV for secure containers by 96.1%, achieving 65.7% and 75.4% reductions in the average and 99th percentile end-to-end startup time.

The emerging integrated sensing and communication (ISAC) technologies show great potential for 5G NR, offering a new wireless-sensing infrastructure paradigm. Users can benefit from pervasive sensing applications in various scenarios without communication penalties. Given the diverse demands for sensing resources across different sensing tasks, elastic resource scheduling becomes crucial, particularly when resources are constrained. However, existing approaches often treat users equally, limiting the applicability in dealing with diverse sensing tasks in the real world. In this article, we introduce ElaSe , a pioneering sensing technique that enables elastic and prompt scheduling of sensing resources. At the core of ElaSe is the exploration of the user’s state to precisely determine the sensing resource requirements and schedule resources accordingly. We build the first model for matching sensing resources with sensing demands, and further propose a predictive scheduling scheme to eliminate delays by leveraging the 5G virtualized radio access network (vRAN). ElaSe has been implemented on a CPU-based 5G vRAN and commercial 5G user equipments. We conduct experiments to evaluate the performance of ElaSe under different settings. The results demonstrate that ElaSe outperforms the non-scheduling scheme, with a 34% reduction in trajectory tracking error and a 92% decrease in resource allocation error.

In this paper, we use empirical measurements to show that container network startup is a key factor that contributes to the slow startup of secure containers in multi-tenant clouds, especially in the scenario of serverless computing, where the issue is pronounced by high-volume concurrent container invocations. We conduct extensive and detailed analysis on existing Container Network Interface (CNI) plugins and show that even the fastest one doubles the startup time from the no-network scenario. We show that the major cause of the blowup in total startup time is that enabling networking significantly increases the contention among different startup stages, particularly for global Linux kernel locks, including the Routing Table NetLink (RTNL) mutex lock and various spin locks. We reveal that contending for these locks hinders startup performance in three ways, including directly increasing stage time, causing poor pipeline overlap and wasting CPU resources. To mitigate such kernel lock contention, we propose a multi-stage concurrency control mechanism based on Bayesian optimization to limit the concurrency of each contended stage. Our results show that this lightweight mechanism can effectively reduce the end-to-end container startup time by 18.8% with negligible extra overhead.

Integrated Sensing and Communication (ISAC) has been witnessed to be a new paradigm of wireless sensing in 5G networks. Users can benefit from pervasive sensing applications in various scenarios with no communication penalty. Given the diverse demands for sensing resources across different sensing tasks, elastic resource scheduling becomes crucial, particularly when resources are constrained. However, existing approaches often treat users equally, limiting their applicability in dealing with diverse sensing tasks in the real world. In this paper, we introduce ElaSe, a pioneering sensing technique that enables real-time elastic scheduling of sensing resources. At the core of ElaSa is the exploration of the user's state to precisely determine the sensing resource requirements and schedule resources accordingly. We build the first model for matching sensing resources with sensing demands, and further propose a predictive scheduling scheme to eliminate delays by leveraging the 5G virtualized radio access network (vRAN). We conduct experiments to evaluate the performance of ElaSe under different settings. The results demonstrate that ElaSe outperforms the non-scheduling scheme, with a 34% reduction in trajectory tracking error and a 92% decrease in resource allocation error.

Vibration measurement is a crucial task in industrial systems, where vibration characteristics reflect health conditions and indicate anomalies of the devices. Previous approaches either work in an intrusive manner or fail to capture the micrometer-level vibrations. In this work, we propose mmVib, a practical approach to measure micrometer-level vibrations with mmWave radar. First, we derive a metric calledVibration Signal-to-Noise Ratio(VSNR) that highlights the directions of reducing measurement errors of tiny vibrations. Then, we introduce the design of mmVib based on the concept ofMulti-Signal Consolidation(MSC) for the error reduction and multi-object measurement. We implement a prototype of mmVib, and the experiments show that it achieves$3.946\%$relative amplitude error and$0.02487\%$relative frequency error in median. Typically, the average amplitude error is only$3.174um$when measuring the$100um$-amplitude vibration at around 5 meters. Compared to two existing mmWave-based approaches, mmVib reduces the 80th-percentile amplitude error by$69.21\%$and$97.99\%$respectively.

Recently, mmWave has been widely used in fine-grained sensing applications due to its short wavelength and large bandwidth. One mmWave device usually can measure the target's 1D micro-displacement along the line-of-sight (LOS) direction. In this work, we try to empower mmWave with the capability of measuring 2D micro-displacements. Our insight is that although the mmWave reflection from one path contains only 1D observation, the spatial separability of mmWave offers an opportunity to separate multipath reflections from the received signal. Combining the coherent observations from multipath reflections can restore the 2D orbit of the target. Based on this insight, we present GWaltz, a mmWave sensing system that manages to measure sub-mm-level 2D orbits of rotating machinery. In GWaltz, we first reveal the relationship between the rotor's movement and the observed ghost multipath reflections (GMRs) and then design a set of novel signal processing techniques to restore the rotor orbit from the poor-quality GMR signals. We implement GWaltz with a commercial mmWave radar, and our evaluation results show that it achieves an absolute error of about 8.42um when measuring 100um-diameter rotor orbits.

Vibration measurement is a crucial task in industrial systems, where vibration characteristics reflect the health and indicate anomalies of the objects. Previous approaches either work in an intrusive manner or fail to capture the micrometer-level vibrations. In this work, we propose mmVib, a practical approach to measure micrometer-level vibrations with mmWave radar. By introducing a Multi-Signal Consolidation (MSC) model to describe the properties of the reflected signals, we exploit the inherent consistency among those signals to accurately recover the vibration characteristics. We implement a prototype of mmVib, and the experiments show that this design achieves 8.2% relative amplitude error and 0.5% relative frequency error in median. Typically, the median amplitude error is 3.4um for the 100um-amplitude vibration. Compared to two existing approaches, mmVib reduces the 80th-percentile amplitude error by 62.9% and 68.9% respectively.

Liquid leakage detection is a crucial issue in modern industry, which concerns industrial safety. Traditional solutions, which generally rely on specialized sensors, suffer from intrusive deployment, high cost, and high power consumption. Such problems prohibit applying those solutions for large-scale and continuously industrial monitoring. In this work, we present a RFID-based solution, TwinLeak, to detect liquid leakage using COTS RFID devices. Detecting the leakage accurately with coarse-grained RSSI and phase readings of tags has been a daunting task, which is especially challenging when low detection delay is required. Our system achieves these goals based on the fact that the inductive coupling between two adjacent tags is highly sensitive to the liquid leaked between them. Therefore, instead of judging according to the signals of each individual tag, TwinLeak utilizes the relationship between the signals of two tags as an effective feature for leakage detection. Specifically, Twin-Leak extracts discriminative signal features from short segments of signals and instantly identifies leakage using a light-weight classifier. A model-guided method for leakage progress tracking is further devised to simultaneously estimate the leakage volume and rate. We implement TwinLeak, evaluate its performance across various scenarios, and deploy it in a real-world industrial IoT system. In average, TwinLeak achieves a TPR higher than 97.2%, a FPR lower than 0.5%, and a relative property estimation error around 10%, while triggering early alarms after only about 4.6mL liquid leaks.

RFID tracking has attracted significant interest from both academia and industry due to its low cost and ease of deployment. Previous works focus more on tracking in 2D space or separately consider tracking of the location and the orientation. They especially struggle in 3D situations due to the increase in the degree of freedom and the limited information conveyed by the RFID tags. In this paper, we propose 3D-OmniTrack, an approach that can accurately track the 3D location and orientation of an object. We introduce a polarization-sensitive phase model in an RFID system, which takes into consideration both the distance and the 3D posture of an object. Based on this model, we design an algorithm to accurately track the object in 3D space. We conduct real-world experiments and present results that show 3D-OmniTrack can achieve centimeter-level location accuracy with the average orientation error of 5°. 3D-OmniTrack has significant advantages in both the accuracy and the efficiency, compared with state-of-the-art approaches.

Backscatter communication, due to its low energy consumption, attract a broad range of applications. The throughput of such low-power communication is however limited. Parallel backscatter is deemed as a promising technique for improving the overall throughput by enabling concurrent transmissions of the backscattering tags. The state-of-the-art approaches for parallel backscatter assume that all the states of the collided signals are distinguishable in the In-phase and Quadrature (IQ) signal plane. In this paper, we disclose the superclustering phenomenon that makes the assumption untenable and significantly degrades the overall performance. Moreover, we observe that the indistinguishable states at different channels are not the same due to the intrinsic channel diversity. Motivated by the observation, we propose Canon, an approach that exploits the channel diversity of the backscatter tags for reliable parallel decoding. In Canon, we address two critical challenges: (i) designing the Multi-Carrier Backscatter (MCB) module to extract the collided signals simultaneously from multiple channels, (ii) designing the Multi-Channel Cluster Union (MCCU) algorithm to distinguish each state of the collided signals. The experiments demonstrate that Canon can achieve over 10 times higher throughput than the state-of-the-art approaches.

The development of Internet-of-Things (IoT) technologies in recent years brings us unprecedented opportunities for innovations in the power industry. This demo abstract introduces our research and practice with Pavatar - IoT for the power industry. Pavatar includes a series of system deployments in the core sections of Global Energy Internet (GEI), for the purposes of automatic surveillance and remote diagnosis of ultra-high-voltage converter stations (UHVCSs). Pavatar incorporates technologies like lower-power or battery-free sensing, cross-technology communication, edge computing, machine learning, and enhances the user experience with 3D virtual reality. The deployed system significantly reduces the manpower cost and enhances the operational efficiency of the UHVCS.

Software-defined networking (SDN) is deemed as a promising direction to offer generalizability of wireless sensor networks (WSN). To introduce SDN into WSNs, however, means a series of non-trivial challenges due to the wireless and ad-hoc nature of WSNs. In this paper, we present our study towards a software-defined architecture for multi-function wireless sensor networks. Our proposal called Pangu is built upon the opportunistic routing protocol stack and introduces the concept of modality properties of sensor nodes. It enables centralized network control over a WSN while preserving the flexibility of underlying ad-hoc routing. We tackle the critical problems of the architecture design by presenting three essential components of Pangu. Moreover, we implement Pangu on a real-world testbed and evaluate it with various experiments.