Quality, Reliability, Security and Robustness in Heterogeneous Systems

With robotics technologies advancing rapidly, there are many new robotics applications such as surveillance, mining tasks, search and rescue, and autonomous armies. In this work, we focus on use of robots for target searching. For example, a collection of Unmanned Aerial Vehicle (UAV) could be sent to search for survivor targets in disaster rescue missions. We assume that there are multiple targets. The moving speeds and directions of the targets are unknown. Our objective is to minimize the searching latency which is critical in search and rescue applications. Our basic idea is to partition the search area into grid cells and apply the divide-and-conquer approach. We propose two searching strategies, namely, the circuit strategy and the rebound strategy. The robots search the cells in a Hamiltonian circuit in the circuit strategy while they backtrack in the rebound strategy. We prove that the expected searching latency of the circuit strategy for a moving target is upper bounded by \(\frac{3n^2-4n+3}{2n}\) where n is the number of grid cells of the search region. In case of a static or suerfast target, we derive the expected searching latency of the two strategies. Simulations are conducted and the results show that the circuit strategy outperforms the rebound strategy.

Modern sensor networks work on the basis of intelligent sensors and actuators, their connection is carried out using conventional or specifically dedicated networks. The efficiency and smooth transmission of such a network is of great importance for the accuracy of measurements, sensor energy savings, or transmission speed. Ethernet in many networks is typically based on the TCP/IP protocol suite. Regardless of whether or not the network transmission is wired or wireless, it should always be reliable. TCP ensures transmission reliability through retransmissions, congestion control and flow control. But TPC is different in networks based on the UDP protocol. The most important here is the transmission speed achieved by shortening the header or the lack of an acknowledgment mechanism. Assuming the network is an automatic control system, it has interconnected elements that interact with each other to perform some specific tasks such as speed control, reliability and security of transmission, just the attributes that define stability being one of the fundamental features of control systems. Such a system returns to equilibrium after being unbalanced. There are many definitions of stability, e.g. Laplace or Lupanov. To check the stability of the sensor network connected to the Internet, different stability criteria should be used. We are going to analyze the stability of a computer network as a dynamic linear system, described by the equations known in the literature. In this paper, we propose the method of testing stability for positive systems using the Metzlner matrix in sensor networks such as IoT or IIoT. We will carry out tests in a place where wide area networks connect to sensor networks, that is in gates.

Mobile Crowd Sensing is an emerging paradigm, which engages ordinary mobile device users to efficiently collect data and share sensed information using mobile applications. The data collection of participants consumes computing, storage and communication resources; thus, it is necessary to give rewards to users who contribute their private data for sensing tasks. Furthermore, since the budget of the sensing task is limited, the Service Provider (SP) needs to select a set of participants such that the total utility of their sensing data can be maximized, and their bid price for sensing data can be satisfied without exceeding the total budget. In this paper, firstly, we claim that the total data utility of a set of participants within a certain area should be calculated according to the data quality of each participant and the location coverage of the sensing data. Secondly, a participant selection scheme has been proposed, which determines a set of participants with maximum total data utility under the budget constraint, and shows that it is a Quadratic Integer Programming problem. Simulations have been conducted to solve the selection problem. The Simulation results demonstrate the effectiveness of the proposed scheme.

Task offloading emerges as a promising solution in Mobile Edge Computing (MEC) scenarios to not only incorporate more processing capability but also save energy. There however exists a key conflict between the heavy processing workloads of terminals and the limited wireless bandwidth, making it challenging to determine the computing placement at the terminals or the remote servers. In this paper, we aim to migrate the most suitable offloading tasks to fully obtain the benefits from the resourceful cloud. The problem in this task offloading scenario is modeled as an optimization problem. Therefore, a Genetic Algorithm is then proposed to achieve maximal user selection and the most valuable task offloading. Specifically, the cloud is pondered to provide computing services for as many edge wireless terminals as possible under the limited wireless channels. The base stations (BSs) serve as the edge for task coordination. The tasks are jointly considered to minimize the computing overhead and energy consumption, where the cost model of local devices is used as one of the optimization objectives in this wireless mobile selective schedule. We also establish the multi-devices task offloading scenario to further verify the efficiency of the proposed allocating schedule. Our extensive numerical experiments demonstrate that our allocating scheme can effectively take advantage of the cloud server and reduce the cost of end users.

Mobile crowdsensing (MCS) depends on mobile users to collect sensing data, whose quality highly depends on the expertise/experience of the users. It is critical for MCS to identify right persons for a given sensing task. A commonly-used strategy is to “teach-before-use”, i.e., training users with a set of questions and selecting a subset of users who have answered the questions correctly the most of times. This method has large room for improvement if we consider users’ learning curve during the training process. As such, we propose an interactive learning pattern recognition framework, Goldilocks, that can filter users based on their learning patterns. Goldilocks uses an adaptive teaching method tailored for each user to maximize her learning performance. At the same time, the teaching process is also the selecting process. A user can thus be safely excluded as early as possible from the MCS tasks later on if her performance still does not match the desired learning pattern after the training period. Experiments on real-world datasets show that compared to the baseline methods, Goldilocks can identify suitable users to obtain more accurate and more stable results for multi-categories classification problems.

As the core infrastructure of cloud computing, the datacenter networks place heavy demands on efficient storage and management of massive data. Data placement strategy, which decides how to assign data to nodes for storage, has a significant impact on the performance of the datacenter. However, most of the existing solutions cannot be better adaptive to the dynamics of the network. Moreover, they focus on where to store the data (i.e., the selection of storage node) but have not considered how to store them (i.e., the selection of routing path). Since reinforcement learning (RL) has been developed as a promising solution to address dynamic network issues, in this paper, we integrate RL into the datacenter networks to deal with the data placement issue. Considering the dynamics of resources, we propose a Q-learning based data placement strategy for datacenter networks. By leveraging Q-learning, each node can adaptively select next-hop based on the network information collected from downstream, and forward the data toward the storage node that has adequate capacity along the path with high available bandwidth. We evaluate our proposal on the NS-3 simulator in terms of average delay, throughput, and load balance. Simulation results show that the Q-learning placement strategy can effectively reduce network delay and increase average throughout while achieving load-balanced among servers.

Physical machines in modern datacenters are routinely upgraded due to their maintenance requirements, which involves migrating all the virtual machines they currently host to alternative physical machines. For this kind of datacenter upgrades, it is critical to minimize the time it takes to upgrade all the physical machines in the datacenter, so as to reduce disruptions to cloud services. To minimize the upgrade time, it is essential to carefully schedule the migration of virtual machines on each physical machine during its upgrade, without violating any constraints imposed by virtual machines that are currently running. Rather than resorting to heuristic algorithms, we propose a new scheduler, Raven, that uses an experience-driven approach with deep reinforcement learning to schedule the virtual machine migration process. With our design of the state space, action space and reward function, Raven trains a fully-connected neural network using the cross-entropy method to approximate the policy of a choosing destination physical machine for each migrating virtual machine. We compare Raven with state-of-the-art heuristic algorithms in the literature, and our results show that Raven effectively leads to shorter time to complete the datacenter upgrade process.

Cloud containers represent a new, light-weight alternative to virtual machines in cloud computing. A user job may be described by a container graph that specifies the resource profile of each container and container dependence relations. This work is the first in the cloud computing literature that designs efficient market mechanisms for container based cloud jobs. Our design targets simultaneously incentive compatibility, computational efficiency, and economic efficiency. It further adapts the idea of batch online optimization into the paradigm of mechanism design, leveraging agile creation of cloud containers and exploiting delay tolerance of elastic cloud jobs. The new and classic techniques we employ include: (i) compact exponential optimization for expressing and handling non-traditional constraints that arise from container dependence and job deadlines; (ii) the primal-dual schema for designing efficient approximation algorithms for social welfare maximization; and (iii) posted price mechanisms for batch decision making and truthful payment design. Theoretical analysis and trace-driven empirical evaluation verify the efficacy of our container auction algorithms.

With the rapid development of information technology, it has become increasingly popular to handle and analyze complex relationships of various information network applications, such as social networks and biological networks. An unsolved primary challenge is to find a way to represent the network structure to efficiently compute, process and analyze network tasks. Graph Neural Network (GNN) based node representation learning is an emerging learning paradigm that embeds network nodes into a low dimensional vector space through preserving the network topology as possible. However, existing GNN architectures have limitation in distinguishing the position of nodes with the similar topology, which is crucial for many network prediction and classification tasks. Anchors are defined as special nodes which are in the important positions, and carries a lot of interactive information with other normal nodes. In this paper, we propose Anchors-aware Graph Neural Networks (A-GNN), which can make the vectors of node embedding contain location information by introducing anchors. A-GNN first selects the set of anchors, computes the distance of any given target node to each anchor, and afterwards learns a non-linear distance-weighted aggregation scheme over the anchors. Therefore A-GNN can obtain global position information of nodes regarding the anchors. A-GNN are applied to multiple prediction tasks including link prediction and node classification. Experimental results show that our model is superior to other GNN architectures on six datasets, in terms of the ROC, AUC accuracy score.

In recent years, with the rapid growth of big data and computation, high-performance computing and heterogeneous computing have been widely concerned. In object detection algorithms, people tend to pay less attention to training time, but more attention to algorithm running time, energy efficiency ratio and processing delay. FPGA can achieve data parallel operation, low power, low latency and reprogramming, providing powerful computing power and enough flexibility. In this paper, SDAccel tool of Xilinx is used to implement a heterogeneous computing platform for face detection based on CPU+FPGA, in which FPGA is used as a coprocessor to accelerate face detection algorithm. A high-level synthesis (HLS) approach allows developers to focus more on the architecture of the design and lowers the development threshold for software developers. The implementation of Viola Jones face detection algorithm on FPGA is taken as an example to demonstrate the development process of SDAccel, and explore the potential parallelism of the algorithm, as well as how to optimize the hardware circuit with high-level language. Our final design is 70 times faster than a single-threaded CPU.

In this paper, we consider a cooperative uplink network consisting of two users, a half-duplex decode-and-forward (DF) relay and a base station (BS). In the relaying network, the two users transmit packets to the buffer-aided relay using non-orthogonal multiple access (NOMA) or orthogonal multiple access (OMA) technology. We proposed a hybrid NOMA/OMA based mode selection (MS) scheme, which adaptively switches between the NOMA and OMA transmission modes according to the instantaneous strength of wireless links and the buffer state. Then, the state transmission matrix probabilities of the corresponding Markov chain is analyzed, and the performance in terms of sum throughput, outage probability, average packet delay and diversity gain are evaluated with closed form expressions. Numerical results are provided to demonstrate that hybrid NOMA/OMA achieves significant performance gains compared to conventional NOMA and OMA in most scenarios.

Acoustic communication has become a research focus without requiring extra hardware on the receiver side and facilitates numerous near-field applications such as mobile payment, data sharing. To communicate, existing researches either use audible frequency band or inaudible one. The former gains a high throughput but endures being audible, which can be annoying to users. The latter, although inaudible, falls short in throughput due to the limited available (near) ultrasonic bandwidth (18–22 kHz). In this paper, we achieve both high speed and inaudibility for acoustic communication by modulating the coded acoustic signal (0–20 kHz) on ultrasonic carrier. By utilizing the nonlinearity effect on microphone, the modulated audible acoustic signal can be demodulated and then decoded. We design and implement UltraComm, an inaudible acoustic communication system with OFDM scheme based on the characteristics of the nonlinear speaker-to-microphone channel. We evaluate UltraComm on different mobile devices and achieve throughput as high as 16.24 kbps, meanwhile, keep inaudibility.

Millimeter-wave network based on beamforming is an interference-limited network. In order to mitigate the interference for the 5G millimeter-wave cellular network, the concept of cooperative multi-beam transmission (Beam-CoMP) is proposed in this paper to improve cell capacity. For users in the beam overlapping zone, there is strong interference between beams, so for such users, overlapping beams provide services to users through cooperation. This method can solve the problems of poor edge coverage and serious interference of overlapping coverage of beams at the same time. The specific process of beam cooperation is given and the Beam-CoMP method proposed is simulated to verify its effectiveness in improving the UE performance. The results show that cell capacity increases with the increase of the number of users in the service beam.

Path validation has long be explored toward forwarding reliability of Internet traffic. Adding cryptographic primitives in packet headers, path validation enables routers to enforce which path a packet should follow and to verify whether the packet has followed the path. How to implement path validation for multi-path routing is yet to be investigated. We find that it leads to an impractically low efficiency when simply applying existing single-path validation to multi-path routing.

In this paper, we present ProMPV as an initiative to explore efficient multi-path validation for multi-path routing. We segment the forwarding path into segments of three routers following a sliding window with size one. Based on this observation, we design ProMPV as a proactive multi-path validation protocol in that it requires a router to proactively leave to its second next hop with proofs that cannot be tampered by its next hop. In multi-path routing, this greatly optimizes the computation and packet size. A packet no longer needs to carry all proofs of routers along all paths. Instead, it iteratively updates its carried proofs that correspond to only three hops. We validate the security and performance of ProMPV through security analysis and experiment results, respectively.

Advanced metering infrastructure (AMI) is the core component of the smart grid. As the wireless connection between smart meters in AMI is featured with high packet loss and low transmission rate, AMI is considered as a representative of the low power and lossy networks (LLNs). In such communication environment, the routing protocol in AMI network is essential to ensure the reliability and real-time of data transmission. The IPv6 routing protocol for low-power and lossy networks (RPL), proposed by IETF ROLL working group, is considered to be the best routing solution for the AMI communication environment. However, the performance of RPL can be seriously degraded due to jamming attack. In this paper, we analyze the performance degradation problem of RPL protocol under jamming attack. We propose a backup node selection mechanism based on the standard RPL protocol. The proposed mechanism chooses a predefined number of backup nodes that maximize the probability of successful transmission. We evaluation the proposed mechanism through MATLAB simulations, results show the proposed mechanism improves the performance of RPL under jamming attack prominently.

The opportunistic network adopts the disconnected store-and-forward architecture to provide communication support for the nodes without an infrastructure. As there is no stable communication link between the nodes, so that forwarding messages is via any encountered nodes. Social networks based on such opportunistic networks will have privacy challenges. In this paper, we propose a privacy protection scheme routing based on the utility value. We exploit the Bloom filter to obfuscate the friends lists and the corresponding utility values of nodes in order to make the routing decisions. This is easy to implement with high performance. Considering no infrastructure and stable link in opportunistic networks, this paper presents a self-organized key management system consisting of an identity authentication scheme based on the zero-knowledge proof of the elliptic curve and a key agreement scheme based on the threshold cryptography. The nodes prove their identities by themselves, and each node carries a certificate library to improve the authentication efficiency and success rate. In order to ensure the forward security and improve the session key agreement rate and the success rate, we exploit threshold cryptography to divide the session key, which could reduce the communication consumption of the traditional Diffie-Hellman (DH) algorithm. The experimental simulation results show that the proposed schemes are much better than the existing schemes for opportunistic networks.