In this paper, we present GDA, a generalized decision aggregation framework that integrates information from distributed sensor nodes for decision making in a resource efficient manner. Different from traditional approaches, our proposed GDA framework is able to not only estimate the reliability of each sensor, but also take advantage of its confidence information, and thus achieves higher decision accuracy. Targeting generalized problem domains, our framework can naturally handle the scenarios where different sensor nodes observe different sets of events whose numbers of possible classes may also be different. GDA also makes no assumption about the availability level of ground truth label information, while being able to take advantage of any if present. For these reasons, our approach can be applied to a much broader spectrum of sensing scenarios. In this paper, we also propose two extensions of the GDA framework, i.e., incremental GDA (I-GDA) and parallel GDA (P-GDA) to deal with streaming and large-scale data. The advantages of our proposed methods are demonstrated through both theoretic analysis and extensive experiments.

Due to the rapid growth of real-time applications and the ubiquity of IEEE 802.11 MAC as a layer-2 protocol for wireless local area networks (WLANs), it becomes increasingly important to support delay-based quality of service (QoS) in such WLANs. In this paper, we develop a simple and accurate enough analytical model for predicting the queueing delay of real-time multimedia traffic flows in non-homogeneous random access based WLANs. This leads to tractable analysis for meeting queueing delay specifications of a number of flows. In particular, we address the feasibility problem of whether the mean delays required by a set of User Datagram Protocol (UDP) flows supporting real-time multimedia traffic can be guaranteed in WLANs. Based on the model and feasibility analysis, we further develop an optimization technique to minimize the delays for the traffic flows. Moreover, we present a decentralized algorithm and report its implementation and present extensive simulation and experimental trace-based results to demonstrate the accuracy of our model and the performance of the algorithms.

In this paper, we present GDA, a generalized decision aggregation framework that integrates information from distributed sensor nodes for decision making in a resource efficient manner. Traditional approaches that target similar problems only take as input the discrete label information from individual sensors that observe the same events. Different from them, our proposed GDA framework is able to take advantage of the confidence information of each sensor about its decision, and thus achieves higher decision accuracy. Targeting generalized problem domains, our framework can naturally handle the scenarios where different sensor nodes observe different sets of events whose numbers of possible classes may also be different. GDA also makes no assumption about the availability level of ground truth label information, while being able to take advantage of any if present. For these reasons, our approach can be applied to a much broader spectrum of sensing scenarios. The advantages of our proposed framework are demonstrated through both theoretic analysis and extensive experiments.

Internet-scale data centers (IDCs) have rapidly proliferated to such an extent that their energy consumption and GreenHouse Gas (GHG) emissions have become an important concern to society. As a result, many IDC operators have started using renewable energy, e.g., wind power, to power their data centers. Unfortunately, the utilization of wind energy has stayed at a low ratio due to the intermittent nature of wind. This paper makes the case that it is in fact possible for a distributed IDC system to exploit multiple uncorrelated wind energy sources to significantly reduce the effect of intermittency and nearly achieve “entirely green” cloud-scale services. This result is obtained based on the analysis of real-world wind power traces from 69 wind farms. The idea is to leverage the front-end load dispatching server to send work to the location where wind power is available. We propose a wind-power-aware (WPA) policy that routes jobs based only on the current states of workloads and wind power availabilities in the data centers. We show that with the WPA policy more than 95% of energy consumption in IDCs can in fact be satisfied by wind power, and, secondly, that achieving this does not require the delaying of processing of jobs due to wind availability. We also show that the locations where data centers are placed play an important role in achieving high wind power utilization. Our analysis shows that wind power utilization can generally lie in a range from 44% to 96%, depending on how the locations of wind farms are selected. We propose a method for location selection that uses the coefficient of variation instead of the correlation coefficient, and show that with this method the utilization can lie in the high end of the above range. Finally, we verify these results by simulations that are based on real-world traces for both workloads and wind power generations.

Mobile devices adopt the IEEE 802.11 PSM (Power-Saving Mode) scheme and its enhancements to reduce their energy consumption when using Wi-Fi interfaces. However, the capability of PSM to save energy is limited when the WLANs are highly congested by other Wi-Fi clients. In this paper, instead of further pursuing the trade-off between power saving and the incurred delay on the client side, we take a different approach and explore the energy saving potential by considering the scheduling policy on the Access Point (AP) side. We find that the traditional packet-level first-come-first-serve policy is not sleep optimal since it keeps the PSM clients awake unnecessarily. We propose SOFA, an AP-centric scheme, which helps PSM clients save energy by minimizing the time they are forced to stay awake while down link traffic is being transmitted to other clients. SOFA delivers down link packets to the PSM clients in an optimal sequence, such that several objective are simultaneously achieved: (i) system-sleep optimality, (ii) energy-fairness, (iii) attention fairness, and (iv) no unnecessary deferral of packets beyond a beacon period. First, it determines an attention quota for each client at the beginning of each beacon period, without requiring any knowledge of available wireless capacity. Then it takes the attention "quota'' and attention request as inputs to decide the down link packet scheduling. We prove the stability and optimality of SOFA. Simulation results shows SOFA dramatically decreases the energy consumption of PSM clients in a crowded WLAN, especially for those clients with small attention requests.

Due to the rapid growth of real-time applications and the ubiquity of IEEE 802.11 MAC as a layer-2 protocol for wireless local area networks (WLANs), it is of increasing interest to support quality of service (QoS) in such WLANs. In this paper, we develop a simple but accurate enough analytical model for predicting queueing delay in non-homogeneous random access based WLANs. This leads to tractable solutions for meeting queueing delay specifications of a number of flows. Using this model, we address the feasibility problem of whether the mean delays required by a set of inelastic flows can be guaranteed in WLANs. Based on the model and feasibility analysis, we further develop an optimization technique to minimize the delays for inelastic flows. We present extensive simulation results to demonstrate the accuracy of our model and the performance of the algorithms.

IEEE 802.11 DCF is the dominant protocol used in existing WLANs. However, the efficiency of DCF progressively degrades with the increase of contending clients in the network as well as the wireless link rate. To address this issue, in this paper, we present a distributed random media access protocol, named CHAIN, which significantly improves uplink performance of WLANs. CHAIN mainly uses overhearing to coordinate clients in a network, and thus introduces little control overhead. The key in CHAIN is a novel piggyback transmission opportunity. In CHAIN, clients maintain a precedence relation among one another, and a client can immediately transmit a new packet after it overhears a successful transmission of its predecessor, without going through the regular contending process. When the network load is low, CHAIN behaves similar to DCF; But when the network becomes congested, clients automatically start chains of transmissions to improve efficiency. CHAIN is derived from DCF and co-exists friendly with it. Moreover, it possesses all the advantages of the 802.11 DCF standard - simplicity, robustness, and scalability. We analytically prove the correctness and fairness of CHAIN. Our extensive simulations on J-SIM verify our analytical results, and demonstrate significant performance gain of CHAIN over DCF.

This paper aims at achieving optimal rate allocation for data aggregation in wireless sensor networks. We first formulate this rate allocation problem as a network utility maximization problem. Due to its non-convexity, we take a couple of variable substitutions on the original problem and transform it into an approximate problem, which is convex. We then apply duality theory to decompose this approximate problem into a rate control subproblem and a scheduling subproblem. Based on this decomposition, a distributed algorithm for joint rate control and scheduling is designed, and proved to approach arbitrarily close to the optimum of the approximate problem. Finally, we show that our approximate solution can achieve near-optimal performance through both theoretical analysis and simulations.

In wireless networks, how to select transmit power that maximizes throughput is a challenging problem. On one hand, transmissions at a high power level could increase interference to others; on the other hand, transmissions at a low power level are prone to being interfered by others. Prior works consider this problem as a search for afixedoptimal power setting that maximizes communication spatial reuse. In this paper, we pursue a novel approach that combines power selection with a random medium access mechanism. For each transmission, a noderandomlyselects a transmit power from all available power levels to access the medium. In this way, all combinations of network power settings could be selected with some probability. Using a recently developed Markov chain model, we derive a distributed scheme that determines the access probabilities of each power setting, according to the arrival rate of traffic and the service rate achieved by the scheme. We show that this scheme always converges to the optimal solution. Moreover, we also show that the random scheme can attain the maximal throughput region that can be obtained by any time-sharing between power settings, and which is consequently larger than the region any fixed power setting can achieve.

Solar-powered sensor nodes have incentive to spend extra energy, especially when the battery is fully charged, because this energy surplus would be wasted otherwise. In this paper, we consider the problem of utilizing such energy surplus to adaptively adjust the redundancy level of erasure codes used in communication, so that the delivery reliability is improved while the network lifetime is still conserved. We formulate the problem as maximizing the end-to-end packet delivery probability under energy constraints. This formulated problem is hard to solve because of the combinatorics involved and the special curvature of its objective function. By exploiting its inherent properties, we propose an effective solution called SolarCode, which has a constant approximation ratio. We evaluate SolarCode in the context of our solar-powered sensor network testbed. Experiments show that SolarCode is successful in utilizing energy surplus and leads to higher data delivery reliability.

With emerging demand for online surveillance and management, persistent and reliable connectivity to base stations (BSs) is indispensable so that network operators, which have access to all BSs, can collect measurements from all wireless terminals in the field, monitor their status and respond to critical incidents in real-time. However, disconnected networks make reliable BS connectivity problematic. Many papers have studied the problem of placing the minimum number of relays to retain connectivity. However, none of them considered polymorphous networks, which have multiple topologies, due to terminal movement, unsynchronized wakeup schedule and packet forwarding policy. Our paper makes the first attempt towards the study of relay placement problem in polymorphous networks. We propose two heuristic algorithms, topology stitch algorithm and topology iterative algorithm, which are built upon constrained relay placement algorithms for monomorphous networks with a single topology. Also, we propose the weigh-and-place algorithm (WPA), which optimizes relay placement across topologies with balanced load based on integer programming formulation. Evaluation shows that WPA places a smaller number of relays than the other two heuristic algorithms and achieves good load balance among multiple BSs.

In WiFi-based content distribution community infrastructure (CDCI), file servers are deployed in diverse locations around cities, caching popular files interesting to a community. They serve file download requests from mobile users in proximity via WiFi. In this paper, we study the optimal caching strategy for file servers in CDCI so that the mean file retrieval probability within deadline is maximized, subject to storage capacity constraints of each file server. The optimal caching strategy depends on many factors, such as users' mobility patterns, access point topology, file popularity, etc. We have formalized this content management problem as a mixed integer optimization problem. Because of NPhardness and a large variable space of this optimization problem, we propose a heuristic algorithm MobaSsign to allocate file blocks. Extensive experiments show that our heuristic caching strategy considering mobility patterns improves the file retrieval probability within deadline.

In this paper we study the issue of topology control under the physical signal-to-interference-noise-ratio (SINR) model, with the objective of maximizing network capacity. We show that existing graph-model-based topology control captures interference inadequately under the physical SINR model, and as a result, the interference in the topology thus induced is high and the network capacity attained is low. Towards bridging this gap, we propose a centralized approach, called spatial reuse maximizer (MaxSR), that combines a power control algorithm T2P with a topology control algorithm P2T. T2P optimizes the assignment of transmit power given a fixed topology, where by optimality we mean that the transmit power is so assigned that it minimizes the average interference degree (defined as the number of interfering nodes that may interfere with the ongoing transmission on a link) in the topology. P2T, on the other hand, constructs, based on the power assignment made in T2P, a new topology by deriving a spanning tree that gives the minimal interference degree. By alternately invoking the two algorithms, the power assignment quickly converges to an operational point that maximizes the network capacity. We formally prove the convergence of MaxSR. We also show via simulation that the topology induced by MaxSR outperforms that derived from existing topology control algorithms by 50%-110% in terms of maximizing the network capacity.