INFOCOM 2021 论文列表

Anomaly detection is indispensable in modern IT infrastructure management. However, the dimension explosion problem of the monitoring data (large-scale machines, many key performance indicators, and frequent monitoring queries) causes a scalability issue to the existing algorithms. We propose a coarse-to-fine model transfer based framework CTF to achieve a scalable and accurate data-center-scale anomaly detection. CTF pre-trains a coarse-grained model, uses the model to extract and compress per-machine features to a distribution, clusters machines according to the distribution, and conducts model transfer to fine-tune per-cluster models for high accuracy. The framework takes advantage of clustering on the per-machine latent representation distribution, reusing the pre-trained model, and partial-layer model fine-tuning to boost the whole training efficiency. We also justify design choices such as the clustering algorithm and distance algorithm to achieve the best accuracy. We prototype CTF and experiment on production data to show its scalability and accuracy. We also release a labeling tool for multivariate time series and a labeled dataset to the research community.

With the increasing adoption of private blockchain platforms, consortia operating in various sectors such as trade, finance, logistics, etc., are becoming common. Despite having the benefits of a completely decentralized architecture which supports transparency and distributed control, existing private blockchains limit the data, assets, and processes within its closed boundary, which restricts secure and verifiable service provisioning to the end-consumers. Thus, platforms such as e-commerce with multiple sellers or cloud federation with a collection of cloud service providers cannot be decentralized with the existing blockchain platforms. This paper proposes a decentralized gateway architecture interfacing private blockchain with end-users by leveraging the unique combination of public and private blockchain platforms through interoperation. Through the use case of decentralized cloud federations, we have demonstrated the viability of the solution. Our testbed implementation with Ethereum and Hyperledger Fabric, with three service providers, shows that such consortium can operate within an acceptable response latency while scaling up to 64 parallel requests per second for cloud infrastructure provisioning. Further analysis over the Mininet emulation platform indicates that the platform can scale well with minimal impact over the latency as the number of participating service providers increases.

Mobile sensing is an effective way to collect environmental data such as air quality, humidity and temperature at low costs. However, mobile robots are typically battery powered and have limited travel distances. To accelerate data collection in large geographical areas, it is beneficial to deploy multiple robots to perform tasks in parallel. In this paper, we investigate the Multi-Robot Informative Path Planning (MIPP) problem, namely, to plan the most informative paths in a target area subject to the budget constraints of multiple robots. We develop two deep reinforcement learning (RL) based cooperative strategies: independent learning through credit assignment and sequential rollout based learning for MIPP. Both strategies are highly scalable with the number of robots. Extensive experiments are conducted to evaluate the performance of the proposed and baseline approaches using real-world WiFi Received Signal Strength (RSS) data. In most cases, the RL based solutions achieve superior or similar performance as a baseline genetic algorithm (GA)-based solution but at only a fraction of running time during inference. Furthermore, when the budgets and initial positions of the robots change, the pre-trained policies can be applied directly.

With the advancement of deep neural networks and computer vision-based Human Activity Recognition, employment of Point-Cloud Data technologies (LiDAR, mmWave) has seen a lot interests due to its privacy preserving nature. Given the high promise of accurate PCD technologies, we develop, PALMAR, a multiple-inhabitant activity recognition system by employing efficient signal processing and novel machine learning techniques to track individual person towards developing an adaptive multi-inhabitant tracking and HAR system. More specifically, we propose (i) a voxelized feature representation-based real-time PCD fine-tuning method, (ii) efficient clustering (DBSCAN and BIRCH), Adaptive Order Hidden Markov Model based multi-person tracking and crossover ambiguity reduction techniques and (iii) novel adaptive deep learning-based domain adaptation technique to improve the accuracy of HAR in presence of data scarcity and diversity (device, location and population diversity). We experimentally evaluate our framework and systems using (i) a real-time PCD collected by three devices (3D LiDAR and 79 GHz mmWave) from 6 participants, (ii) one publicly available 3D LiDAR activity data (28 participants) and (iii) an embedded hardware prototype system which provided promising HAR performances in multi-inhabitants (96%) scenario with a 63% improvement of multi-person tracking than state-of-art framework without losing significant system performances in the edge computing device.

Network-consensus-based decentralized learning optimization algorithms have attracted a significant amount of attention in recent years due to their rapidly growing applications. However, most of the existing decentralized learning algorithms could not achieve low sample and communication complexities simultaneously - two important metrics in evaluating the trade-off between computation and communication costs of decentralized learning. To overcome these limitations, in this paper, we propose a triple hybrid decentralized stochastic gradient descent (TH-DSGD) algorithm for efficiently solving non-convex network-consensus optimization problems for decentralized learning. We show that to reach an ϵ2-stationary solution, the total sample complexity of TH-DSGD is O(ϵ-3) and the communication complexity is O(ϵ-3), both of which are independent of dataset sizes and significantly improve the sample and communication complexities of the existing works. We conduct extensive experiments with a variety of learning models to verify our theoretical findings. We also show that our TH-DSGD algorithm is stable as the network topology gets sparse and enjoys better convergence in the large-system regime.

We present an extensive experimental evaluation of the performance and power consumption of the 60 GHz IEEE 802.11ad technology on commercial smartphones. We also compare 802.11ad against its main competitors in the 5 GHz band - 802.11ac and, for first time, 802.11ax, on mobile devices. Our performance comparison focuses on two aspects that have not been extensively studied before: (i) dense multi-client and multi-AP topologies and (ii) popular mobile applications under realistic mobility patterns. Our power consumption study covers both non-communicating and communicating modes. We also present the first study of the power saving mode in 802.11ad-enabled smartphones and its impact on performance. Our results show that 802.11ad is better able to address the needs of emerging bandwidth-intensive applications in smartphones than its 5 GHz counterparts. At the same time, we identify several key research directions towards realizing its full potential.

RDMA is increasingly deployed in data center to meet the demands of ultra-low latency, high throughput and low CPU overhead. However, it is not easy to migrate existing applications from the TCP/IP stack to the RDMA. The developers usually need to carefully select communication primitives and manually tune the parameters for each single-purpose system. After operating the high-speed RDMA network, we identify multiple hidden costs which may cause degraded and/or unpredictable performance of RDMA-based applications. We demonstrate these hidden costs including the combination of complicated parameter settings, scalability of Reliable Connections, two-sided memory management and page alignment, resource contention among diverse traffics, etc. Furthermore, to address these problems, we introduce Nem, a suite that allows developers to maximize the benefit of RDMA by i) eliminating the resource contention at NIC cache through asynchronous resource sharing; ii) introducing hybrid page management based on messages sizes; iii) isolating flows of different traffic classes based hardware features. We implement the prototype of Nem and verify its effectiveness by rebuilding the RPC message service, which demonstrates the high throughput for large messages, low latency for small messages without compromising the low CPU utilization and good scalability performance for a large number of active connections.

A large number of consensus algorithms have been proposed. However, the requirement of strict consistency limits their wide adoption, especially in high-performance required systems. In this paper, we propose a weak consensus algorithm that only maintains the consistency of relative positions between the messages. We apply this consensus algorithm to construct a high-performance blockchain system, called Sphinx. We implement the system with 32k+ lines of code including all components like consensus/P2P/ledger/etc. The evaluations show that Sphinx can reach a peak throughput of 43k TPS (with 8 full nodes), which is significantly faster than current blockchain systems such as Ethereum given the same experimental environment. To the best of our knowledge, we present the first weak consensus algorithm with a fully implemented blockchain system.

As one of the recently emerged threats to Deep Learning (DL) models, clean-label data poisoning can teach DL models to make wrong predictions on specific target data, such as images or network traffic packets, by injecting a small set of poisoning data with clean labels into the training datasets. Although several clean-label poisoning methods have been developed before, they have two main limitations. First, the methods developed with bi-level optimization or influence functions usually require second-order information, leading to substantial computational overhead. Second, the methods based on feature collision are not very transferable to unseen feature spaces or generalizable to various scenarios. To address these limitations, we propose a first-order efficient general-purpose clean-label poisoning attack in this paper. In our attack, we first identify the first-order model update that can push the model towards predicting the target data as the attack targeted label. We then formulate a necessary condition based on the model update and other first-order information to optimize the poisoning data. Theoretically, we prove that our first-order poisoning method is an approximation of a second-order approach with theoretically-guaranteed performance. Empirically, extensive evaluations on image classification and network traffic classification demonstrate the outstanding efficiency, transferability, and generalizability of our poisoning method.

Wi-Fi scanning that searches neighboring access points (APs) is an essential prerequisite for Wi-Fi operations such as initial association and handover. As the traffic demand increases, APs are more densely deployed and the number of operating Wi-Fi channels also increases, which, however, results in additional scanning delay and makes the scanning a burdensome task. In this paper, we note that the co-location of Wi-Fi protocol with BLE protocol is a common practice in IoT networks, and develop a Wi-Fi passive scanning framework that uses BLE to assist scanning. Although the framework has great potential to improve scanning performance without explicit message exchanges, there are technical challenges related to time synchronization and channel switching delay. We address the challenges and develop a practical passive scanning scheme, named BLESS-Sync. We verify its performance through testbed experiments and extensive simulations, and show that BLESS-Sync significantly outperforms legacy Wi-Fi scanning in terms of scanning delay and energy efficiency.

Switch buffer serves an important role in modern internet. To achieve efficiency, today's switches often use on-chip shared memory. Shared memory switches rely on buffer management policies to allocate buffer among ports. To avoid waste of buffer resources or a few ports occupy too much buffer, existing policies tend to maximize overall buffer utilization and pursue queue length fairness. However, blind pursuit of utilization and misleading fairness definition based on queue length leads to buffer occupation with no benefit to throughput but extends queuing delay and undermines burst absorption of other ports. We contend that a buffer management policy should proactively detect port traffic and adjust buffer allocation accordingly. In this paper, we propose Traffic-aware Dynamic Threshold (TDT) policy. On the basis of classic dynamic threshold policy, TDT proactively raise or lower port threshold to absorb burst traffic or evacuate meaningless buffer occupation. We present detailed designs of port control state transition and state decision module that detect real time traffic and change port thresholds accordingly. Simulation and DPDK-based real testbed demonstrate that TDT simultaneously optimizes for throughput, loss and delay, and reduces up to 50% flow completion time.

URLLC is an important use case in 5G NR that targets at 1-ms level delay-sensitive applications. For fast transmission of URLLC traffic, a promising mechanism is to multiplex URLLC traffic into a channel occupied by eMBB service through preemptive puncturing. Although preemptive puncturing can offer transmission resource to URLLC on demand, it will adversely affect throughput and link reliability performance of eMBB service. To mitigate such an adverse impact, a possible approach is to employ link adaptation (LA) through MCS selection for eMBB users. In this paper, we study the problem of maximizing eMBB throughput through MCS selection while ensuring link reliability requirement for eMBB users. We present DELUXE - the first successful design and implementation based on deep learning to address this problem. DELUXE involves a novel mapping method to compress high-dimensional eMBB transmission information into a low-dimensional representation with minimal information loss, a learning method to learn and predict the block-error rate (BLER) under each MCS, and a fast calibration method to compensate errors in BLER predictions. For proof of concept, we implement DELUXE through a link-level 5G NR simulator. Extensive experimental results show that DELUXE can successfully maintain the desired link reliability for eMBB while striving for spectral efficiency. In addition, our implementation can meet the real-time requirement (<; 125 μs) in 5G NR.

Rule update in TCAM has long been identified as a key technical challenge due to the rule order constraint. Existing algorithms take each rule update as an independent task. However, emerging applications produce batch rule update requests. Processing the updates individually causes high aggregated cost which can strain the processor and/or incur excessive TCAM lookup interrupts. This paper presents the first true batch update algorithm, ABUT. Unlike the other alleged batch update algorithms, ABUT collectively evaluates and optimizes the TCAM placement for whole batches throughout. By applying the topology grouping and maintaining the group order invariance in TCAM, ABUT achieves substantial computing time reduction yet still yields the best-in-class placement cost. Our evaluations show that ABUT is ideal for low-latency and high-throughput batch TCAM updates in modern high-performance switches.

Backscatter networks are becoming a promising solution for embedded sensing. In these networks, backscatter sensors are deeply implanted inside objects or living beings and form a deep backscatter network (DBN). The fundamental challenges in DBNs are the significant attenuation of the wireless signal caused by environmental materials (e.g., water and bodily tissues) and the miniature antennas of the implantable backscatter sensors, which prevent existing backscatter networks from powering sensors beyond superficial depths. This study presents RiCharge, a turbocharging solution that enables powering up and communicating with DBNs through a single augmented RF source, which allows existing backscatter sensors to serve DBNs at zero startup cost. The key contribution of RiCharge is the turbocharging algorithm that utilizes RF surges to induce constructive power surges at deep backscatter sensors in accordance with the FCC regulations, for overcoming the turn-on voltage barrier. RiCharge is implemented in commodity devices, and the evaluation result reveals that RiCharge can use only a single RF source to power up backscatter sensors at 60 m distance in the air (i.e., 10x longer than a commercial off-the-shelf reader) and 50 cm-depth under water (i.e., 2x deeper than the previous record).

Erasure coding has been extensively deployed in today's data centers to tackle prevalent failures, yet it is prone to give rise to substantial cross-rack traffic for parity update. In this paper, we propose a new rack-coordinated update mechanism to suppress the cross-rack update traffic, which comprises two successive phases: a delta-collecting phase that collects data delta chunks, and another selective parity update phase that renews the parity chunks based on the update pattern and parity layout. We further design RackCU, an optimal rack-coordinated update solution that achieves the theoretical lower bound of the cross-rack update traffic. We finally conduct extensive evaluations, in terms of large-scale simulation and real-world data center experiments, showing that RackCU can reduce 22.1%-75.1% of the cross-rack update traffic and hence improve 34.2%-292.6% of the update throughput.

This paper presents a network layer model for a wireless multiple access system with both persistent and nonpersistent users. There is a single access point with multiple identical channels. Each user who wants to send a file first scans a subset of the channels to find one that is idle. If at least one idle channel is found, the user transmits a file over that channel. If no idle channel is found, a persistent user will repeat the access attempt at a later time, while a nonpersistent user will leave. This is a useful mathematical model for situations where a group of persistent users stay near an access point for an extended period of time while nonpersistent users come and go. Users have heterogeneous activity behavior, file upload rates, and service durations. The system is a complex multi-dimensional Markov chain. The steady state probabilities are found by exploiting a latent reversibility property and leveraging a discrete Fourier transform. This enables simple expressions for throughput and blocking probability.

Shadow banning consists for an online social net-work in limiting the visibility of some of its users, without them being aware of it. Twitter declares that it does not use such a practice, sometimes arguing about the occurrence of "bugs" to justify restrictions on some users. This paper is the first to address the plausibility of shadow banning on a major online platform, by adopting both a statistical and a graph topological approach.We first conduct an extensive data collection and analysis campaign, gathering occurrences of visibility limitations on user profiles (we crawl more than 2.5 millions of them). In such a black-box observation setup, we highlight the salient user profile features that may explain a banning practice (using machine learning predictors). We then pose two hypotheses for the phenomenon: i) limitations are bugs, as claimed by Twitter, and ii) shadow banning propagates as an epidemic on user-interaction ego-graphs. We show that hypothesis i) is statistically unlikely with regards to the data we collected. We then show some interesting correlation with hypothesis ii), suggesting that the interaction topology is a good indicator of the presence of groups of shadow banned users on the service.

In this paper, we adopt the fluid limits to analyze Age of Information (AoI) in a wireless multiaccess network with many users. We consider the case wherein users have heterogeneous i.i.d. channel conditions and the statuses are generate-at-will. Convergence of the AoI occupancy measure to the fluid limit, represented by a Partial Derivative Equation (PDE), is proved within an approximation error inversely proportional to the number of users. Global convergence to the equilibrium of the PDE, i.e., stationary AoI distribution, is also proved. Based on this framework, it is shown that an existing AoI lower bound in the literature is in fact asymptotically tight, and a simple threshold policy, with the thresholds explicitly derived, achieves the optimum asymptotically. The proposed threshold-based policy is also much easier to decentralize than the widely-known index-based policies which require comparing user indices. To showcase the usability of the framework, we also use it to analyze the average non-linear AoI functions (with power and logarithm forms) in wireless networks. Again, explicit optimal threshold-based policies are derived, and average age functions proven. Simulation results show that even when the number of users is limited, e.g., 10, the proposed policy and analysis are still effective.

The current Internet content delivery model assumes strict mapping between a resource and its descriptor, e.g., a JPEG file and its URL. Content Distribution Networks (CDNs) extend it by replicating the same resources across multiple locations, and introducing multiple descriptors. The goal of this work is to build Web-LEGO, an opt-in service, to speedup webpages at client side. Our rationale is to replace the slow original content with fast similar or equal content. Further, we perform a reality check of this idea both in term of the prevalence of CDN-less websites, availability of similar content, and user perception of similar webpages via millions of scale automated tests and thousands of real users. Then, we devise Web-LEGO, and address natural concerns on content inconsistency and copyright infringements. The final evaluation shows that Web-LEGO brings significant improvements both in term of reduced Page Load Time (PLT) and user-perceived PLT. Specifically, CDN-less websites provide more room for speedup than CDN-hosted ones, i.e., 7x more in the median case. Besides, Web-LEGO achieves high visual accuracy (94.2%) and high scores from a paid survey: 92% of the feedback collected from 1,000 people confirm Web-LEGO's accuracy as well as positive interest in the service.