INFOCOM 2019 论文列表

Vehicular Cloud Computing (VCC) is a new technological shift which exploits the computation and storage resources on vehicles for computational service provisioning. Spare onboard resources are pooled by a VCC operator, e.g. a roadside unit, to serve computational tasks using the vehicle-as-a-resource framework. This paper investigates timely service provisioning for deadline-constrained tasks in VCC systems by leveraging the task replication technique (i.e., allowing one task to be executed by vehicles). A learning-based algorithm, called DATEV (Deadline-Aware Task rEplication for Vehicular Cloud), is proposed to address the special issues in VCC systems including uncertainty of vehicle movements, volatile vehicle members, and large vehicle population. The proposed algorithm is developed based on a novel contextual-combinatorial multi-armed bandit learning framework. DATE-V is “ contextual ” because it utilizes side information (context) of vehicles and tasks to infer the completion probability of a task replication under random vehicle movements. DATE-V is “combinatorial” because it replicates the received task and sends task replications to multiple vehicles to guarantee the service timeliness. When learning with multi-armed bandit, DATE-V also addresses the practical concern of delayed feedbacks caused by the task transmission/computational delay in using VCC. We rigorously prove that our learning algorithm achieves a sublinear regret bound compared to an oracle algorithm that knows the exact completion probability of any task replications. Simulations are carried out based on real-world vehicle movement traces and the results show that DATE-V significantly outperforms benchmark solutions.

Evolved Multimedia Broadcast Multicast Service (eMBMS) is a 3GPP standard that improves the utilization of scarce wireless resources and the quality of the received content. eMBMS uses a Single Frequency Network (SFN) to transmit real-time videos over synchronized resources across neighboring base stations (eNBs) and allows users to share wireless spectrum across multiple cell sites. However the user with the worst channel condition and the eNB with the least available resources limit the throughput of a session. To overcome such limitations, the SFN can be divided into non-overlapping clusters of eNBs and in each cluster users can be split into groups. We formulate an optimization problem that maximizes an operator-defined utility for multiple eMBMS sessions served at multiple bitrates by choosing the optimal set of SFN clusters and user groups for each session. We propose an algorithm, RTOP, that finds the optimal or a near-optimal solution in real-time regardless of the number of eMBMS users. Our extensive simulations indicate that, in comparison to state-of-the-art schemes, RTOP improves the system utility and average user bitrate by up to 14% and 90% respectively. Additionally, we show that the utility of RTOP always stays within a 1% gap from the optimal solution.

Having multiple data stores that can potentially serve content is common in modern networked applications. Data stores often publish approximate summaries of their content to enable effective utilization. Since these summaries are not entirely accurate, forming an efficient access strategy to multiple data stores becomes a complex risk management problem.This paper formally models this problem, and introduces practical algorithms with guaranteed approximation ratios, and in particular we show that our algorithms are optimal in a variety of settings. We also perform an extensive simulation study based on real data, and show that our algorithms are more robust than existing heuristics. That is, they exhibit near optimal performance in various settings whereas the efficiency of existing approaches depends upon system parameters that may change over time, or be otherwise unknown.

Production machine learning (ML) clusters commonly host a variety of distributed ML workloads, e.g., speech recognition, machine translation. While server sharing among jobs improves resource utilization, interference among co-located ML jobs can lead to significant performance downgrade. Existing cluster schedulers (e.g., Mesos) are interference-oblivious in their job placement, causing suboptimal resource efficiency. Interference-aware job placement has been studied in the literature, but was treated using detailed workload profiling and interference modeling, which is not a general solution. This paper presents Harmony, a deep learning-driven ML cluster scheduler that places training jobs in a manner that minimizes interference and maximizes performance (i.e., training completion time). Harmony is based on a carefully designed deep reinforcement learning (DRL) framework augmented with reward modeling. The DRL employs state-of-the-art techniques to stabilize training and improve convergence, including actor-critic algorithm, job-aware action space exploration and experience replay. In view of a common lack of reward samples corresponding to different placement decisions, we build an auxiliary reward prediction model, which is trained using historical samples and used for producing reward for unseen placement. Experiments using real ML workloads in a Kubernetes cluster of 6 GPU servers show that Harmony outperforms representative schedulers by 25% in terms of average job completion time.

Segment Routing (SR) is a powerful tool to solve traffic engineering in large networks. It enables steering the traffic along any arbitrary network path while limiting scalability issues as routers do not need to maintain a global state. Mathematical programming approaches proposed so far for SR either do not scale well with the size of topology or impose a strong limit on the number of possible detours (typically at most one). Moreover they do not support Segment Routing fully by ignoring the adjacency segments. This paper leverages column generation, a widely used technique for solving large scale linear programs, combined with a novel dynamic program for solving the pricing problem. Our approach reaches near optimal solutions with gap guarantees by also computing a strong lower-bound tighter than the multi-commodity flow relaxation. It scales even on large topologies and exploits the full expressiveness of SR including adjacency segments. Our experiments compared with existing traffic engineering techniques on various topologies and demand matrices demonstrate the advantages of our approach in terms of scalability, any-time behavior and quality of the solutions.

The use of in-band full duplex (IBFD) is a promising improvement over classical TDD or FDD communication schemes. To enable IBFD radios, the electrical balance duplexer (EBD) has been proposed to suppress the direct self-interference (SI) at the RF stage. The remaining SI is typically assumed to be canceled further in the digital domain. In this paper, we show that the non-zero Doppler frequencies can be extracted from the residual SI, giving information about the speed of objects in the environment. As a result, an IBFD radio can be seen as a monostatic Doppler radar which is affected by the communication signal. This paper presents a detailed performance analysis of the different sources of interference affecting the Doppler radar. The performance is also evaluated using a radar-enhanced IBFD prototype consisting of a SDR module and an EBD designed to achieve up to 55 dB Tx-Rx isolation in the 1.74 GHz RF band. A measurable Doppler component is created by moving a 15 dBsm cone at known velocities at 0. 5-1.3 m from the prototype. For an EBD SI rejection of 45 dB, speeds in the range of 200 to 800 mm/s are detected with high accuracy.

With the recent boom in the cryptocurrency market, hackers have been on the lookout to find novel ways of commandeering users' machine for covert and stealthy mining operations. In an attempt to expose such under-the-hood practices, this paper explores the issue of browser cryptojacking, whereby miners are secretly deployed inside browser code without the knowledge of the user. To this end, we analyze the top 50k websites from Alexa and find a noticeable percentage of sites that are indulging in this exploitative exercise often using heavily obfuscated code. Furthermore, mining prevention plug-ins, such as NoMiner, fail to flag such cleverly concealed instances. Hence, we propose a machine learning solution based on hardware-assisted profiling of browser code in real-time. A fine-grained micro-architectural footprint allows us to classify mining applications with >99% accuracy and even flags them if the mining code has been heavily obfuscated or encrypted. We build our own browser extension and show that it outperforms other plug-ins. The proposed design has negligible overhead on the user's machine and works for all standard off-the-shelf CPUs.

The proliferation of innovative mobile services such as augmented reality, networked gaming, and autonomous driving has spurred a growing need for low-latency access to computing resources that cannot be met solely by existing centralized cloud systems. Mobile Edge Computing (MEC) is expected to be an effective solution to meet the demand for low-latency services by enabling the execution of computing tasks at the network-periphery, in proximity to end-users. While a number of recent studies have addressed the problem of determining the execution of service tasks and the routing of user requests to corresponding edge servers, the focus has primarily been on the efficient utilization of computing resources, neglecting the fact that non-trivial amounts of data need to be stored to enable service execution, and that many emerging services exhibit asymmetric bandwidth requirements. To fill this gap, we study the joint optimization of service placement and request routing in MEC-enabled multi-cell networks with multidimensional (storage-computation-communication) constraints. We show that this problem generalizes several problems in literature and propose an algorithm that achieves close-to-optimal performance using randomized rounding. Evaluation results demonstrate that our approach can effectively utilize the available resources to maximize the number of requests served by low-latency edge cloud servers.

The explosion of 5G networks and the Internet of Things will result in an exceptionally crowded RF environment, where techniques such as spectrum sharing and dynamic spectrum access will become essential components of the wireless communication process. In this vision, wireless devices must be able to (i) learn to autonomously extract knowledge from the spectrum on-the-fly; and (ii) react in real time to the inferred spectrum knowledge by appropriately changing communication parameters, including frequency band, symbol modulation, coding rate, among others. Traditional CPU-based machine learning suffers from high latency, and requires application-specific and computationally-intensive feature extraction/selection algorithms. Conversely, deep learning allows the analysis of massive amount of unprocessed spectrum data without ad-hoc feature extraction. So far, deep learning has been used for offline wireless spectrum analysis only. Therefore, additional research is needed to design systems that bring deep learning algorithms directly on the device's hardware and tightly intertwined with the RF components to enable real-time spectrum-driven decision-making at the physical layer. In this paper, we present RFLearn, the first system enabling spectrum knowledge extraction from unprocessed I/Q samples by deep learning directly in the RF loop. RFLearn provides (i) a complete hardware/software architecture where the CPU, radio transceiver and learning/actuation circuits are tightly connected for maximum performance; and (ii) a learning circuit design framework where the latency vs. hardware resource consumption trade-off can explored. We implement and evaluate the performance of RFLearn on custom software-defined radio built on a system-on-chip (SoC) ZYNQ-7000 device mounting AD9361 radio transceivers and VERT2450 antennas. We showcase the capabilities of RFLearn by applying it to solving the fundamental problems of modulation and OFDM parameter recognition. Experimental results reveal that RFLearn decreases latency and power by about 17x and 15x with respect to a software-based solution, with a comparatively low hardware resource consumption.

This paper presents RF-Mehndi, a passive commercial RFID tag array formed identifier. The key RF-Mehndi novelty is that when the user's fingertip touching on the tag array surface during the communication, the backscattered signals by the tag array become user-dependent and unique. Hence, if we enhance the communication modality of many personal cards nowadays by RF-Mehndi, in case that a card gets lost or stolen, it cannot be used illegally by the adversaries. To harvest such a benefit, we have two key observations in designing RF-Mehndi. The first observation is when tags are nearby, their interrogated currents can change each other's circuit characteristics, based on which unique phase features can be obtained from backscattered signals. The second observation is that when the user's fingertip touches the tag array surface during communication, the phase feature can be further profiled by this user. Based on these observations, the card and its holder can be potentially authenticated at the same time. To transfer the RF-Mehndi idea to a practical system, we further address technical challenges. We implement a prototype system. Extensive evaluations show the effectiveness of RF-Mehndi, achieving excellent authentication performance.

Increasing threats of malicious eavesdropping raise concerns in confidential data reporting by body-worn sensors. We propose a secure, body-guided transmission channel through the use of galvanic coupling (GC). This method involves injecting weak electrical current into the body, which propagates primarily through the skin. The proposed approach makes the transmission of biometric data impervious to sniffing attacks, enabling the body to serve as a waveguide. This paper makes the following contributions: (i) An analytical channel model using a tissue equivalent circuit of the human arm-wrist-palm GC-propagation path is formulated and empirically verified. (ii) A simulation study is conducted for a comparative analysis of various modulation schemes, leveraging the validated GC-channel behavior. (iii) A GC-transceiver with optimized communication parameters (modulation, frequency, power) is designed and implemented using a dielectrically equivalent tissue phantom, and (iv) through experimental trials, resilience to over-the-air susceptibility (i.e., likelihood of adversarial eavesdropping) of the GC-signal and similar body communication techniques are demonstrated. Performance results of the GC-transceiver prototype yield a bit error rate of 10-6with a transmit power of -2dBm, in addition to over 7x reduction of signal radiation outside the body compared to capacitive coupling.

Traffic anomaly detection is critical for advanced Internet management. Existing detection algorithms usually work off-line and cannot timely detect anomalies. They also suffer from high cost for storage and computation. Although online and accurate traffic anomaly detection is very important, it very difficult to achieve. We propose to utilize tensor model to well exploit the multi-dimensional information hidden in the traffic data for more accurate online Internet anomaly detection. We decouple the tensor recovery problem to iteratively solve two sub problems, a tensor factorization sub-problem and an anomaly detection sub-problem. To reduce the high cost for computation and storage involved in tensor factorization, we propose two lightweight techniques to effectively derive factor matrices of tensor in the current window and iteration, taking advantage of tensor decomposition results of the previous window and iteration. We have done extensive experiments using two real traffic traces to compare with three tensor based algorithms and three matrix based algorithms. The experiment results demonstrate that our online anomaly detection algorithm can achieve the same anomaly detection accuracy as that of the best offline tensor based algorithm, but at 6100 times faster speed and with very low storage cost.

Inference on location data has generated a great amount of interest and there are two data sources. GPS location data from GPS-based endpoint devices have high precision, but suffer from intermittent availability, small coverage of users and extra demand on devices' battery. Telecom mobility data, on the other hand, have the complementary advantages of continuous availability and complete coverage but generally with coarse location accuracy. The focus of this paper is on the development of a location insight system for activity labeling and user segment inference, based only on Telecom mobility data enriched with point-of-interest (POI) data. Specifically, we estimate activity patterns based on Bayesian techniques while infer user segments via a tree-based classification algorithm. We test the performance of our inference system via simulated mobility data and investigate the impact of important factors such as granularity of data. We conclude that imprecise location data, enriched with other source of information, can be used for the development of new generation activity inference models.

Cloud database provides an advantageous platform for outsourcing of database service. To protect data confidentiality from an untrusted cloud, the original database is often encrypted and then uploaded to the cloud. However, in order to support functional queries, existing secure databases require users to encrypt their data under the same public/symmetric key, which restricts the usage scenarios since users do not really trust each other in practice. Imagine a scenario where a user uploaded his/her own encrypted data to the cloud database and another user wants to execute private range queries on this data. This scenario occurs in many cases of collaborative statistical analysis where the data provider and analyst are different entities. Then either the data provider must reveal its encryption key or the analyst must reveal the private queries. In this paper, we overcome this restriction for secure range queries by enabling query executions on the multi-key encryption data. We propose a secure cloud database supporting range queries under multiple keys, in which all users could preserve the confidentiality of their own different keys, and do not have to share them with each other. At a higher level, our system is constructed on a two-cloud architecture and a novel distributed two-trapdoor public key cryptosystem. We prove that the proposed scheme achieves the goal of a secure query without leaking data privacy, query privacy, and data access patterns. Finally, we use extensive experiments over a real-world dataset on a commercial cloud platform to verify the efficacy of our proposed scheme.

The combination of recent emerging technologies such as network function virtualization (NFV) and network programmability (SDN) gave birth to the Network Slicing revolution. 5G networks consist of multi-tenant infrastructures capable of offering leased network “slices” to new customers (e.g., vertical industries) enabling a new telecom business model: Slice-as-a-Service (SlaaS). In this paper, we aim i) to study the slicing admission control problem by means of a multi-queuing system for heterogeneous tenant requests, ii) to derive its statistical behavior model, and iii) to provide a utility-based admission control optimization. Our results analyze the capability of the proposed SlaaS system to be approximately Markovian and evaluate its performance as compared to legacy solutions.

Today's Android systems do not allow users to manage the permissions granted to applications (apps) in a flexible and dynamic way. Recent studies show that apps often misuse these permissions to access private information, or have trapdoors via which other malicious apps can do the same. In this paper, we develop a framework Figment, which consists of set of libraries that developers can easily use to build in fine-grained dynamic permission management capabilities. The users of their apps can readily invoke these capabilities during execution. The apps would potentially run with reduced functionalities if the user does not wish to allow certain permissions. Figment also allows either the developer or a user to specify context aware permissions, which cause different permissions to be granted to the app in different functional modes (contexts). We believe that Figment reduces the attack surface exposed to potentially malicious apps and offers a significant step in preserving user privacy. While the rudimentary version of Figment uses aspect-oriented programming and does not need rooting of the phone or changes to the Android sub-system, we also provide an optional root-level fail safe implementation that facilitates the embedding of dynamic permission management functions in old applications not built by using Figment libraries. We show that Figment offers significant benefits over the Android Marshmallow permission management system with lower runtime overheads; the main penalty is a one time higher compilation overhead.

This paper presents BatchQR, a mobile APP that can batch read the densely arranged QR codes attached to caps of the tubes and vials in clinical and biological labs. The basic idea of BatchQR is to detect each code in the image and then decode in an one-by-one manner. However, the unique characteristics of the QR code and the application scenario bring technical challenges: First, off-the-shelf lightweight object detection mechanisms are unable to distinguish those densely arranged codes that are highly similar to each other; second, the focus area of the camera is limited, which blurs or distorts parts of the image. To this end, we propose a lightweight code detection mechanism, which can adaptively adjust operating parameters to identify densely arranged QR codes in practice. We also propose a simple but effective image refocus mechanism, which takes an auto-focused image and multiple refocused ones, and then replaces the blurred or distorted code parts with the high-quality counterparts in the refocused images. Comprehensive experimental results show that BatchQR can read 160-180 Version 1-L QR codes in batch with 90%-95% accuracy in 10-14s, which is only 4% of the time consumed by the regular QR code reader in the same situation.

In this paper, we propose an application-aware intelligent load balancing system for high-throughput, distributed computing, and data-intensive science workflows. We leverage emerging deep learning techniques for time-series modeling to develop an application-aware predictive analytics system for accurately forecasting GridFTP connection loads. Our solution integrates with a major U.S. CMS Tier-2 site; we use a real dataset representing 670 million GridFTP transfer connections measured over 18 months to drive our predictive analytics solution. First, we perform extensive analysis on this dataset and use the connection loads as an example to study the temporal dependencies between various user-roles and workflow memberships. We use the analysis to motivate the design of a gated recurrent unit (GRU) based deep recurrent neural network (RNN) for modeling long-term temporal dependencies and predicting connection loads. We develop a novel application-aware, predictive and intelligent load balancer, APRIL, that effectively integrates application metadata and load forecast information to maximize server utilization. We conduct extensive experiments to evaluate the performance of our deep RNN predictive analytics system and compare it with other approaches such as ARIMA and multi-layer perceptron (MLP) predictors. The results show that our forecasting model, depending on the user-role, performs between 5.88%-92.6% better than the alternatives. We also demonstrate the effectiveness of APRIL by comparing it with the load balancing capabilities of an existing production Linux Virtual Server (LVS) cluster. Our approach improves server utilization, on an average, between 0.5 to 11 times, when compared with its LVS counterpart.

Based on network function virtualization (NFV) and software defined network (SDN), network slicing is proposed as a new paradigm for building service-customized 5G network. In each network slice, service-required virtual network functions (VNFs) can be flexibly deployed in an on-demand manner, which will support a variety of 5G use cases. However, due to the diverse performance requirements among different 5G scenarios, an adaptive VNF placement approach is needed to automatically accommodate to service-specific requirements. In this paper, we tackle the VNF placement problem by first proposing a general 5G network slice framework, which jointly contains both edge cloud and core cloud servers. Specially, based on the fact that VNF consolidation may cause severe performance degradation, we adopt a demand-supply model to quantity the VNF interference. With an aim to maximize the total throughput of accepted requests, we propose an Adaptive Interference-Aware (AIA) heuristic approach to automatically place VNFs in 5G service-customized network slices. Through simulations on two typical 5G scenarios, we demonstrate that AIA can efficiently handle traffic variation especially caused by VNF interference and improve the total throughput by 20.11% and 24.21% in autonomous driving and 4K/8K HD video network slices as compared with the state-of-the-art methods.