INFOCOM 2023 论文列表

Although billions of battery-free backscatter devices (e.g., RFID tags) are intensively deployed nowadays, they are still unsatisfying in performance limitations (i.e., short reading range and high miss-reading rate) resulting from power harvesting inefficiency. However, applying classic beamforming technique to backscatter systems meets the deadlock start problem, i.e., without enough power, the backscatter cannot wake up to provide channel parameters; but, without channel parameters, the system cannot form beams to provide power. In this work, we propose a new beamforming paradigm called transfer beamforming (TBF), namely, beamforming strategies can be transferred from reference tags with known positions to power up unknown neighbor tags of interest. Transfer beamforming (is accomplished) via (launching) beamforming (to reference tags firstly) for (the purpose of) transfer. To do so, we adopt semi-active tags as reference tags, which can be easily powered up with a normal reader. Then beamforming is initiated and transferred to power up passive tags surrounded by reference tags. A prototype evaluation of TBF with 8 antennas presents a 99.9% inventory coverage rate in a crowded warehouse with 2,160 RFID tags. Our evaluation reveals that TBF improves the power transmission by 6.9 dB and boosts the inventory speed by 2 × compared with state-of-art methods.

Large models have attracted much attention in the AI area. The sparsely activated mixture-of-experts (MoE) technique pushes the model size to a trillion-level with a sub-linear increase of computations as an MoE layer can be equipped with many separate experts, but only one or two experts need to be trained for each input data. However, the feature of dynamically activating experts of MoE introduces extensive communications in distributed training. In this work, we propose PipeMoE to adaptively pipeline the communications and computations in MoE to maximally hide the communication time. Specifically, we first identify the root reason why a higher pipeline degree does not always achieve better performance in training MoE models. Then we formulate an optimization problem that aims to minimize the training iteration time. To solve this problem, we build performance models for computation and communication tasks in MoE and develop an optimal solution to determine the pipeline degree such that the iteration time is minimal. We conduct extensive experiments with 174 typical MoE layers and two real-world NLP models on a 64-GPU cluster. Experimental results show that our PipeMoE almost always chooses the best pipeline degree and outperforms state-of-the-art MoE training systems by 5%-77% in training time.

Federated learning (FL) is an emerging paradigm of distributed machine learning. However, when applied to wireless network scenarios, FL usually suffers from high communication cost because clients need to transmit their updated gradients to a server in every training round. Although many gradient compression techniques like sparsification and quantization are proposed, they compress clients’ gradients independently, without considering the correlations among gradients. In this paper, we propose SVDFed, a collaborative gradient compression framework for FL. SVDFed utilizes Singular Value Decomposition (SVD) to find a few basis vectors, whose linear combination can well represent clients’ gradients at a certain round. Due to the correlations among gradients, these basis vectors can still well approximate new gradients in many subsequent rounds. With the help of basis vectors, clients only need to upload the coefficients of the linear combination to the server, which greatly reduces communication cost. In addition, SVDFed leverages the classical PID (Proportional, Integral, Derivative) control to determine the proper time to update basis vectors to maintain their representation ability. Through experiments, we demonstrate that SVDFed outperforms existing gradient compression methods in FL. For example, compared to a popular gradient quantization method QSGD, SVDFed can reduce the communication overhead by 66 % and pending time by 99 %.

Federated learning (FL) enables massive clients to collaboratively train a global model by aggregating their local updates without disclosing raw data. Communication has become one of the main bottlenecks that prolongs the training process, especially under large model variances due to skewed data distributions. Existing efforts mainly focus on either single momentum-based gradient descent, or random client selection for potential variance reduction, yet both often lead to poor model accuracy and system efficiency. In this paper, we propose FedMoS, a communication-efficient FL framework with coupled double momentum-based update and adaptive client selection, to jointly mitigate the intrinsic variance. Specifically, FedMoS maintains customized momentum buffers on both server and client sides, which track global and local update directions to alleviate the model discrepancy. Taking momentum results as input, we design an adaptive selection scheme to provide a proper client representation during FL aggregation. By optimally calibrating clients' selection probabilities, we can effectively reduce the sampling variance, while ensuring unbiased aggregation. Through a rigid analysis, we show that FedMoS can attain the theoretically optimal O(T - 2/3) convergence rate. Extensive experiments using real-world datasets further validate the superiority of FedMoS, with 58%-87% communication reduction for achieving the same target performance compared to state-of-the-art techniques.

Emerging reconfigurable datacenters allow to dynamically adjust the network topology in a demand-aware manner. These datacenters rely on optical switches which can be reconfigured to provide direct connectivity between racks, in the form of edge-disjoint matchings. While state-of-the-art optical switches in principle support microsecond reconfigurations, the demand-aware topology optimization constitutes a bottleneck.This paper proposes a dynamic algorithms approach to improve the performance of reconfigurable datacenter networks, by supporting faster reactions to changes in the traffic demand. This approach leverages the temporal locality of traffic patterns in order to update the interconnecting matchings incrementally, rather than recomputing them from scratch. In particular, we present six (batch-)dynamic algorithms and compare them to static ones. We conduct an extensive empirical evaluation on 176 synthetic and 39 real-world traces, and find that dynamic algorithms can both significantly improve the running time and reduce the number of changes to the configuration, especially in networks with high temporal locality, while retaining matching weight.

In a streaming application, an operator is usually instantiated into multiple tasks for parallel processing. Tasks across operators have various memory demands due to different processing logic (e.g., stateful vs. stateless tasks). The memory demands of tasks from the same operator could also vary and fluctuate due to workload variability. Improper memory provision will cause some tasks to have relatively high latency, or even unbound latency that can eventually lead to system instability. We found that the task with the maximum latency of an operator has a significant and even decisive impact on the end-to-end latency.In this paper, we present our task-level memory manager. Based on our quantitative modeling of memory and task-level latency, the manager can adaptively allocate optimal memory size to each task for minimizing the end-to-end latency. We integrate our memory management on Apache Flink. The experiments show that our memory management could significantly reduce end-to-end latency for various applications at different scales and configurations, compared to the Flink native setting.

The enormous amount of data produced by mobile and IoT devices has motivated the development of federated learning (FL), a framework allowing such devices (or clients) to collaboratively train machine learning models without sharing their local data. FL algorithms (like FedAvg) iteratively aggregate model updates computed by clients on their own datasets. Clients may exhibit different levels of participation, often correlated over time and with other clients. This paper presents the first convergence analysis for a FedAvg-like FL algorithm under heterogeneous and correlated client availability. Our analysis highlights how correlation adversely affects the algorithm’s convergence rate and how the aggregation strategy can alleviate this effect at the cost of steering training toward a biased model. Guided by the theoretical analysis, we propose CA-Fed, a new FL algorithm that tries to balance the conflicting goals of maximizing convergence speed and minimizing model bias. To this purpose, CA-Fed dynamically adapts the weight given to each client and may ignore clients with low availability and large correlation. Our experimental results show that CA-Fed achieves higher time-average accuracy and a lower standard deviation than state-of-the-art AdaFed and F3AST, both on synthetic and real datasets.

Streamers are the core competency of the crowd-sourced live streaming (CLS) platform. However, little work has explored how different factors relate to their popularity evolution patterns. In this paper, we will investigate a critical problem, i.e., how to discover the promising streamers in their early stage? To tackle this problem, we first conduct large-scale measurement on a real-world CLS dataset. We find that streamers can indeed be clustered into two evolution types (i.e., rising type and normal type), and these two types of streamers will show differences in some inherent properties. Traditional time-sequential models cannot handle this problem, because they are unable to capture the complicated interactivity and extensive heterogeneity in CLS scenarios. To address their shortcomings, we further propose Niffler, a novel heterogeneous attention temporal graph framework (HATG) for predicting the evolution types of CLS streamers. Specifically, through the graph neural network (GNN) and gated-recurrent-unit (GRU) structure, Niffler can capture both the interactive features and the evolutionary dynamics. Moreover, by integrating the attention mechanism in the model design, Niffler can intelligently preserve the heterogeneity when learning different levels of node representations. We systematically compare Niffler against multiple baselines from different categories, and the experimental results show that our proposed model can achieve the best prediction performance.

Mobile edge computing is becoming one of the ubiquitous computing paradigms to support applications requiring low latency and high computing capability. FPGA-based reconfigurable accelerators have high energy efficiency and low latency compared to general-purpose servers. Therefore, it is natural to incorporate reconfigurable accelerators in mobile edge computing systems. This paper formulates and studies the problem of joint task offloading, access point selection, and resource allocation in heterogeneous edge environments for latency minimization. Due to the heterogeneity in edge computing devices and the coupling between offloading, access point selection, and resource allocation decisions, it is challenging to optimize over them simultaneously. We decomposed the proposed problem into two disjoint subproblems and developed algorithms for them. The first subproblem is to jointly determine offloading and computing resource allocation decisions and is NP-hard, where we developed an algorithm based on semidefinite relaxation. The second subproblem is to jointly determine access point selection and communication resource allocation decisions, where we proposed an algorithm with a provable approximation ratio of 2.62. We conducted extensive numerical simulations to evaluate the proposed algorithms. Results highlighted that the proposed algorithms outperformed baselines and were near-optimal over a wide range of settings.

We consider a wireless network with a base station broadcasting and collecting time-sensitive data to and from spatially distributed nodes in the presence of wireless interference. The Age of Information (AoI) is the time that has elapsed since the most-recently delivered packet was generated, and captures the freshness of information. In the context of broadcast and collection, we define the Age of Broadcast (AoB) to be the amount of time elapsed until all nodes receive a fresh update, and the Age of Collection (AoC) as the amount of time that elapses until the base station receives an update from all nodes. We quantify the average broadcast and collection ages in two scenarios: 1) instance-dependent, in which the locations of all nodes and interferers are known, and 2) instance-independent, in which they are not known but are located randomly, and expected age is characterized with respect to node locations. In the instance-independent case, we show that AoB and AoC scale super-exponentially with respect to the radius of the region surrounding the base station. Simulation results highlight how expected AoB and AoC are affected by network parameters such as network density, medium access probability, and the size of the coverage region.

Distributing quantum entanglements over long distances is essential for the realization of a global scale quantum Internet. Most of the prior work and proposals assume an on-demand distribution of entanglements which may result in significant network resource under-utilization. In this work, we introduce Quantum Overlay Networks (QONs) for efficient entanglement distribution in quantum networks. When the demand to create end-to-end user entanglements is low, QONs can generate and store maximally entangled Bell pairs (EPR pairs) at specific overlay storage nodes of the network. Later, during peak demands, requests can be served by performing entanglement swaps either over a direct path from the network or over a path using the storage nodes. We solve the link entanglement and storage resource allocation problem in such a QON using a centralized optimization framework. We evaluate the performance of our proposed QON architecture over a wide number of network topologies under various settings using extensive simulation experiments. Our results demonstrate that QONs fare well by a factor of 40% with respect to meeting surge and changing demands compared to traditional non-overlay proposals. QONs also show significant improvement in terms of average entanglement request service delay over non-overlay approaches.

Blockchain systems must be secure and offer high performance. These systems rely on transaction broadcast mechanisms to provide both of these features. Unfortunately, in today’s systems, the broadcast mechanisms are highly inefficient.We present Mercury, a new transaction broadcast protocol designed for high performance blockchains. Mercury shortens the transaction propagation delay using two techniques: a virtual coordinate system and an early outburst strategy. Simulation results show that Mercury outperforms prior propagation schemes and decreases overall propagation latency by up to 44%. When implemented in Conflux, an open-source high-throughput blockchain system, Mercury reduces transaction propagation latency by over 50% with less than 5% bandwidth overhead.

With the popularity of LED infrastructure and the camera on smartphone, LED-Camera visible light communication (VLC) has become a realistic and promising technology. However, the existing LED-Camera VLC has limited throughput due to the sampling manner of camera. In this paper, by introducing a polarization dimension, we propose a hybrid modulation scheme with LED and polarization signals to boost throughput. Nevertheless, directly mixing LED and polarized signals may suffer from channel conflict. We exploit well-designed packet structure and Symmetric Return-to-Zero Inverted (SRZI) coding to overcome the conflict. In addition, in the demodulation of hybrid signal, we alleviate the noise caused by polarization on the LED signals by polarization background subtraction. We further propose a pixel-free approach to correct the perspective distortion caused by the shift of view angle by adding polarizers around the liquid crystal array. We build a prototype of this hybrid modulation scheme using off-the-shelf optical components. Extensive experimental results demonstrate that the hybrid modulation scheme can achieve reliable communication, achieving 13.4 kbps throughput, which is 400 % of the existing state-of-the-art LED-Camera VLC.

The cold startup of the container is regarded as a crucial problem to the performance of serverless computing, especially to the resource-capacitated edge clouds. Pre-warming hot containers has been proved as an efficient solution but is at the expense of high memory consumption. Instead of pre-warming a complete container for a function, recent studies advocate Zygote container, which pre-imports some packages and is able to import the other dependent packages at runtime, so as to avoid the cold startup problem. However, as different functions have different package dependencies, how to plan the Zygote generation and pre-warming in a resource-capacitated edge cloud becomes a critical challenge. In this paper, aiming to minimize the overall function startup time and subjective to the resource capacity constraints, we formulate this problem into a Quadratic Integer Programming (QIP) form. We further propose a Randomized Rounding based Zygote Planning (RRZP) algorithm. The performance efficiency of our algorithm is proved via both theoretical analysis and trace-driven simulations. The results show that our algorithm can significantly reduce the startup time by 25.6%.

In-band Network Telemetry (INT) and sketching algorithms are two promising directions for measuring network traffics in real time. To combine sketch with INT and preserve their advantages, a representative approach is to use INT to send a switch sketch in small pieces (called sketchlets) to end-host for reconstructing an identical sketch. However, in this paper, we reveal that when naively selecting buckets to sketchlets, the end-host reconstructed sketch is inaccurate. To overcome this problem, we present DUNE, an innovative sketch-INT network measurement system. DUNE incorporates two key innovations: First, we design a novel scatter sketchlet that is more efficient in transferring measurement data by allowing a switch to select individual buckets to add to sketchlets; Second, we propose lightweight data structures for tracing "freshness" of the sketch buckets, and present algorithms for smartly selecting buckets that contain valuable measurement data to send to end-host. We theoretically prove the effectiveness of our proposed methods, and implement a prototype on commodity programmable switch. The results of extensive experiments driven by real-world traffics on DUNE suggest that our proposed system can substantially improve the measurement accuracy at a trivial cost.

To date, security researchers evaluate their solutions of mitigating denial-of-service (DDoS) attacks via kernel-based or kernel-bypassing testing tools. However, kernel-based tools exhibit poor scalability in attack traffic generation while kernel-bypassing tools result in unacceptable monetary cost. We propose Excalibur, a scalable and low-cost testing framework for DDoS defense solutions. The key idea is to leverage the programmable switch to perform testing tasks with Tbps-level scalability and low cost. Specifically, Excalibur coordinates both a server and a programmable switch to jointly perform testing tasks. It realizes flexible attack traffic generation, which requires a large number of resources, in the server while using the switch to increase the sending rate of attack traffic to Tbps-level. Our experiments on a 64×100Gbps Tofino switch show that Excalibur achieves orders-of-magnitude higher scalability and lower cost than existing tools.

Many Internet platforms collect fresh information of various points of interest (PoIs) relying on users who happen to be nearby the PoIs. The platform will offer reward to incentivize users and compensate their costs incurred from information acquisition. In practice, the user cost (and its distribution) is hidden to the platform, thus it is challenging to determine the optimal reward. In this paper, we investigate how the platform dynamically rewards the users, aiming to jointly reduce the age of information (AoI) and the operational expenditure (OpEx). Due to the hidden cost distribution, this is an online non-convex learning problem with partial feedback. To overcome the challenge, we first design an age-based rewarding scheme, which decouples the OpEx from the unknown cost distribution and enables the platform to accurately control its OpEx. We then take advantage of the age-based rewarding scheme and propose an exponentially discretizing and learning (EDAL) policy for platform operation. We prove that the EDAL policy performs asymptotically as well as the optimal decision (derived based on the cost distribution). Simulation results show that the age-based rewarding scheme protects the platform’s OpEx from the influence of the user characteristics, and verify the asymptotic optimality of the EDAL policy.

The rapid growth of cloud services has brought a significant increase in inter-datacenter traffic. To transfer data among geographically distributed datacenters, cloud providers need to purchase bandwidth from ISPs. The data transferring cost has become one of the major expenses for cloud providers. Therefore, it is essential for a cloud provider to carefully allocate inter-datacenter traffic among the ISPs' links to minimize the costs. Exiting solutions mainly focus on the situations where all links adopt the same pricing scheme. However, in practice, ISPs usually provide multiple pricing schemes for their links due to market competition, which makes the existing solutions nonoptimal. Thus, a new traffic engineering approach that considers various pricing schemes is needed. This paper presents COIN, a new framework for cost-efficient traffic engineering with various pricing schemes. We propose a partition rounding traffic engineering algorithm based on linear independence analysis. The approximation factors and time complexity are formally analyzed. We further conduct large-scale simulations with real- world topologies and datasets. Extensive simulation results show that COIN can save the data transferring cost by up to 54.54% compared with the state-of-the-art solutions.