The growing importance, large scale, and high server density of high-performance computing datacenters make them prone to attacks, misconfigurations, and failures (of the cooling as well as of the computing infrastructure). Such unexpected events often lead to thermal anomalies - hotspots, fugues, and coldspots - which impact the cost of operation of datacenters. A model-based thermal anomaly detection mechanism, which compares expected (obtained using heat-generation and -extraction models) and observedthermal maps (obtained using thermal cameras) of datacenters, is proposed. In addition, a novel Thermal Anomaly-aware Resource Allocation (TARA) is designed to induce a time-varying thermal fingerprint (thermal map) of the datacenter so to maximize the detection accuracy of the anomalies. As shown via experiments on a small-scale testbed as well as via trace-driven simulations, such model-based thermal anomaly detection solution in conjunction with TARA significantly improves the detection probability compared to anomaly detection when scheduling algorithms such as random, round robin, and best-fit-decreasing are employed.

Clouds provide the abstraction of nearly-unlimited computing resources through the elastic use of federated resource pools (virtualized datacenters). They are being increasingly considered for HPC applications, which have traditionally targeted grids and supercomputing clusters. However, maximizing energy efficiency and utilization of cloud datacenter resources, avoiding undesired thermal hotspots (due to overheating of over-utilized computing equipment), and ensuring quality of service guarantees for HPC applications are all conflicting objectives, which require joint consideration of multiple pairwise tradeoffs. An innovative proactive thermal-aware virtual machine consolidation (involving allocations as well as migrations) technique is proposed to maximize computing resource utilization, to minimize data center energy consumption for computing, and to improve the efficiency of heat extraction. The capability to migrate virtual machines away from lightly-loaded servers in a thermal-aware manner opens up opportunity to improve resource consolidation over time and, hence, achieve the aforementioned goals. The effectiveness of the proposed technique is verified through experimental evaluations with HPC workload traces under single- as well as federated-datacenter scenarios.

Innovative mobile applications that rely on real-time in-situ processing of data collected in the fieldneedto tap into the heterogeneous sensing and computing capabilities of sensor nodes, mobile handhelds as well as computing and storage servers in remote datacenters. There is, however,uncertaintyassociated with thequalityandquantityof data from mobile sensors as well as with theavailabilityandcapabilitiesof mobile computing resources on the field. Data and computing-resource uncertainty, if unchecked, may propagate up the “raw data$\rightarrow$information$\rightarrow$knowledge” chain and have an adverse effect on the relevance of the generated results. A generalized workflow representation scheme that can represent a wide variety of data- and task-parallel ubiquitous mobile applications is presented. A unified uncertainty-aware framework for data and computing-resource management to enable real-time, in-situ processing of applications is proposed and evaluated. The framework employs a two-phase solution that captures the propagation of data uncertainty up the data-processing chain using interval arithmetic in the first phase and that employs multi-objective optimization for task allocation in the second phase. The results of a case study to assess effectiveness the proposed framework are discussed in detail. Results reaffirm that i) data-uncertainty awareness helps control the uncertainty in the final result and ii) multi-objective combinatorial approach for task allocation significantly outperforms the single-objective approaches in terms of makespan (15 percent improvement), fairness in battery drain (56 percent improvement), and network load (54 percent improvement).

A data-centric joint adaptive sampling and sleep scheduling solution, SILENCE, for autonomic sensor-based systems that monitor and reconstruct physical or environmental phenomena is proposed. Adaptive sampling and sleep scheduling can help realize the much needed resource efficiency by minimizing the communication and processing overhead in densely deployed autonomic sensor-based systems. The proposed solution exploits the spatiotemporal correlation in sensed data and eliminates redundancy in transmitted data through selective representation without compromising on accuracy of reconstruction of the monitored phenomenon at a remote monitor node. Differently from existing adaptive sampling solutions, SILENCE employs temporal causality analysis to not only track the variation in the underlying phenomenon but also its cause and direction of propagation in the field. The causality analysis and the same correlations are then leveraged for adaptive sleep scheduling aimed at saving energy in wireless sensor networks (WSNs). SILENCE outperforms traditional adaptive sampling solutions as well as the recently proposed compressive sampling techniques. Real experiments were performed on a WSN testbed monitoring temperature and humidity distribution in a rack of servers, and the simulations were performed on TOSSIM, the TinyOS simulator.

Mobile platforms are becoming the predominant medium of access to Internet services due to the tremendous increase in their computation and communication capabilities. However, enabling applications that require real-time in-the-field data collection and processing using mobile platforms is still challenging due to i) the insufficient computing capabilities and unavailability of complete data on individual mobile devices and ii) the prohibitive communication cost and response time involved in offloading data to remote computing resources such as cloud datacenters for centralized computation. A novel resource provisioning framework for organizing the heterogeneous sensing, computing, and communication capabilities of static and mobile devices in the vicinity in order to form an elastic resource pool—a hybrid static/mobile computing grid (also called a loosely-coupled mobile device cloud)—is presented. This local computing grid can be harnessed to enable innovative data- and compute-intensive mobile applications such as ubiquitous context-aware health and wellness monitoring of the elderly, distributed rainfall and flood-risk estimation, distributed object recognition and tracking, and content-based distributed multimedia search and sharing. In order to address challenges such as the inherent uncertainty in the hybrid grid (in terms of network connectivity and device availability), the proposed role-based resource provisioning framework is imparted with autonomic capabilities, namely, self-organization, self-optimization, and self-healing. A thorough experimental analysis aimed at verifying and demonstrating the benefits brought by autonomic capabilities of the framework is also presented in detail.

Due to the rapid growing popularity of mobile Internet, broadband cellular wireless systems are expected to offer higher and higher data rates even in high-mobility environments. Cloud Radio Access Network (C-RAN) is a new centralized paradigm for broadband wireless access that addresses efficiently the fluctuation in capacity demand through real-time inter-Base Station (BS) cooperation. An innovative Blind Source Separation (BSS)-based cellular communication solution for CRANs, Cloud-BSS, which leverages the inter-BS cooperation, is proposed. Cloud-BSS groups contiguous cells into clusters - sets of neighboring cells inside which mobile stations do not need to perform handovers - and allows them to use all of the frequency channels. The proposed solution is studied under different network topologies, and a novel strategy, called Channel-Select, to improve the Signal-to-Noise Ratio (SNR) is introduced. Cloud-BSS enhances the cluster spectral efficiency, decreases handovers, eliminates the need for bandwidth-consuming channel estimation techniques, and mitigates interference. Simulation results, which are discussed along with concepts, confirm these expectations.

Over-the-top estimation of available bandwidth (AB) in a network path has been well studied for wired networks. The AB of a path denotes its slack capacity, i.e., the bandwidth available for use in the path without impacting the existing traffic. This estimation problem has been receiving attention only recently in cellular networks, which are increasingly becoming one of the main modes of access for a large number of applications. In this paper, we discuss the challenges posed by the problem, and why existing techniques developed for wired networks cannot be applied. We show that, interestingly, it may not even be feasible to estimate AB using over-the-top approaches under certain conditions, even when the wireless channel and traffic conditions are non-varying. We then present a novel AB estimation technique for cellular networks, which typically use proportional fair scheduling at base stations. When the wireless channel and traffic conditions are non-varying, our technique can accurately determine AB when it exceeds the “fair share” due to a new flow. We also extend the basic technique for estimation under conditions that are time-varying. The proposed methods can as well be used when one or more bottleneck links in a network path are fair-scheduled using algorithms such as weighted-fair queueing. We evaluate our methods using simulations and over operational networks, and present the results. In simulations, our technique is capable of detecting AB close to 90% of the time under feasible conditions even with bursty traffic.

The growing importance, large scale, and high server density of high-performance computing datacenters make them prone to strategic attacks, misconfigurations, and failures (cooling as well as computing infrastructure). Such unexpected events lead to thermal anomalies - hotspots, fugues, and coldspots - which significantly impact the total cost of operation of datacenters. A model-based thermal anomaly detection mechanism, which compares expected (obtained using heat generation and extraction models) and observed thermal maps (obtained using thermal cameras) of datacenters is proposed. In addition, a Thermal Anomaly-aware Resource Allocation (TARA) scheme is designed to create time-varying thermal fingerprints of the datacenter so to maximize the accuracy and minimize the latency of the aforementioned model-based detection. TARA significantly improves the performance of model-based anomaly detection compared to state-of-the-art resource allocation schemes.

Uncertainties in the wireless communication medium do not allow for guarantees in network performance for cognitive radio applications envisaged for mobile ad hoc emergency networking. The novel concept of mission policies, which specify the Quality of Service (QoS) requirements of the incumbent network as well as the cognitive radio networks, is introduced. The use of mission policies, which vary over time and space, enables graceful degradation in the QoS of incumbent network (only when necessary) based on mission-policy specifications. A Multi-Agent Reinforcement Learning (MARL)-based cross-layer communication framework, RescueNet, is proposed for self-adaptation of nodes in cognitive radio networks. Also, the novel idea of knowledge sharing among the agents (nodes) is introduced to significantly improve the performance of the proposed solution.

The heterogeneous sensing and computing capabilities of sensor nodes, mobile handhelds, as well as computing and storage servers in remote data centers can be harnessed to enable innovative mobile applications that rely on real-time in-situ processing of data generated in the field. There is, however, uncertainty associated with the quality and quantity of data from mobile sensors as well as with the availability and capabilities of mobile computing resources on the field. Data and computing-resource uncertainty, if unchecked, may propagate up the "raw-data→information→knowledge" chain and have an adverse effect on the relevance of the generated results. A unified uncertainty-aware framework for data and computing-resource management is proposed to enable in-situ processing of application workflows on mobile sensing and computing platforms and, hence, to generate actionable knowledge from raw data within realistic time bounds. A two-phase solution that captures the propagation of data-uncertainty up the data-processing chain using interval arithmetic in the first phase and that employs multi-objective optimization for task allocation in the second phase is presented and evaluated in detail.

Clouds provide the abstraction of nearly-unlimited computing resources through the elastic use of federated resource pools (virtualized datacenters). They are being increasingly considered for HPC applications, which have traditionally targeted grids and supercomputing clusters. However, maximizing energy efficiency and utilization of cloud datacenter resources, avoiding undesired thermal hotspots (due to overheating of over-utilized computing equipment), and ensuring quality of service guarantees for HPC applications are all conflicting objectives, which require joint consideration of multiple pairwise tradeoffs. The novel concept of heat imbalance, which captures the unevenness in heat generation and extraction, at different regions inside a HPC cloud datacenter is introduced. This thermal awareness enables proactive datacenter management through prediction of future temperature trends as opposed to the state-of-the-art reactive management based on current temperature measurements. VMAP, an innovative proactive thermal-aware virtual machine consolidation technique is proposed to maximize computing resource utilization, to minimize datacenter energy consumption for computing, and to improve the efficiency of heat extraction. The effectiveness of the proposed technique is verified through experimental evaluations with HPC workload traces under single-as well as federated-datacenter scenarios (in the machine rooms at Rutgers University and University of Florida).