Tsinghua Science and Technology最新文献

The existing multipath routing protocols for wireless sensor networks demonstrate the efficacy of traffic distribution over multiple paths to fulfill the Quality of Service (QoS) requirements of different applications. However, the performance of these protocols is highly affected by the characteristics of the wireless channel and may be even inferior to the performance of single-path approaches. Specifically, when multiple adjacent paths are being used concurrently, the broadcast nature of wireless channels results in interpath interference which significantly degrades end-to-end throughput. In this paper, we propose a Low-Interference Energy-efficient Multipath Routing protocol (LIEMRO) to improve the QoS requirements of event-driven applications. In addition, in order to optimize resource utilization over the established paths, LIEMRO employs a quality-based load balancing algorithm to regulate the amount of traffic injected into the paths. The performance gain of LIEMRO compared to the ETX-based single-path routing protocol is 85%, 80%, and 25% in terms of data delivery ratio, end-to-end throughput, and network lifetime, respectively. Furthermore, the end-to-end latency is improved more than 60%.

Wireless sensor networks are envisioned to be an integral part of cyber-physical systems, yet wireless networks are inherently dynamic and come with various uncertainties. One such uncertainty is wireless communication itself which assumes complex spatial and temporal dynamics. For dependable and predictable performance, therefore, link estimation has become a basic element of wireless network routing. Several approaches using broadcast beacons and/or unicast MAC feedback have been proposed in the past years, but there is still no systematic characterization of the drawbacks and sources of errors in beacon-based link estimation in low-power wireless networks, which leads to ad hoc usage of beacons in routing. Using a testbed of 98 XSM motes (an enhanced version of MICA2 motes), we characterize the negative impact that link layer retransmission and traffic-induced interference have on the accuracy of beacon-based link estimation, and we show that data-driven link estimation and routing achieve higher event reliability (e.g., by up to 18.75%) and transmission efficiency (e.g., by up to a factor of 1.96) than beacon-based approaches. These findings provide solid evidence for the necessity of data-driven link estimation and demonstrate the importance of addressing the drawbacks of beacon-based link estimation when designing protocols for low-power wireless networks of cyber-physical systems.

We study the problem of efficient data aggregation in unreliable wireless sensor networks by designing a fault tolerant data aggregation protocol. A fault tolerant data aggregation protocol consists of two parts: basic aggregation scheduling and amendment strategies. On default, data is aggregated according to the basic aggregation scheduling strategy. The amendment strategy will start automatically when a middle sensor node is out of service. We focus our attention on the amendment strategies and assume that the network adopts a connected dominating set (CDS) based aggregation scheduling as its basic aggregation scheduling strategy. The amendment scheme includes localized aggregation tree repairing algorithms and distributed rescheduling algorithms. The former are used to find a new aggregation tree for every child of the corrupted node, whereas the latter are used to achieve interference free data aggregation scheduling after the amendment. These amendment strategies impact only a very limited number of nodes near the corrupted node and the amendment process is transparent to all the other nodes. Theoretical analyses and simulations show that the scheme greatly improves the efficiency of the data aggregation operation by reducing both message and time costs compared to rebuilding the aggregation tree and rescheduling the entire network.

The performance of distributed computing systems is partially dependent on configuration parameters recorded in configuration files. Evolutionary strategies, with their ability to have a global view of the structural information, have been shown to effectively improve performance. However, most of these methods consume too much measurement time. This paper introduces an ordinal optimization based strategy combined with a back propagation neural network for autotuning of the configuration parameters. The strategy was first proposed in the automation community for complex manufacturing system optimization and is customized here for improving distributed system performance. The method is compared with the covariance matrix algorithm. Tests using a real distributed system with three-tier servers show that the strategy reduces the testing time by 40% on average at a reasonable performance cost.

Modern datacenter and enterprise networks require application identification to enable granular traffic control that either improves data transfer rates or ensures network security. Providing application visibility as a core network function is challenging due to its performance requirements, including high throughput, low memory usage, and high identification accuracy. This paper presents a payload-based application identification method using a signature matching engine utilizing characteristics of the application identification. The solution uses two-stage matching and pre-classification to simultaneously improve the throughput and reduce the memory. Compared to a state-of-the-art common regular expression engine, this matching engine achieves 38% memory use reduction and triples the throughput. In addition, the solution is orthogonal to most existing optimization techniques for regular expression matching, which means it can be leveraged to further increase the performance of other matching algorithms.

FastDu is a file system service that tracks file system changes by intercepting file system calls to maintain directory summaries, which play important roles in both storage administration and improvement of user experiences for some applications. In most circumstances, directory summaries are independently harvested by applications via traversing the file system hierarchy and calling stat () on every file in each directory. For large file systems, this brute-force traverse-based approach can take many hours to complete, even if only a small percentage of the files have changed. This paper describes FastDu, which uses a pre-built database to store harvested directory summaries, and tracks the file system changes by intercepting file system calls, so that new harvesting is restricted to the small subset of directories that contain modified files. Tests using FastDu show that this approach reduces the time needed to get a directory summary by one or two orders of magnitude with almost negligible penalty to application-aware file system performance.

In order to improve micro-CT's capability of accurate quantification of linear attenuation coefficient μ, a dual energy method was developed to correct beam hardening artifacts caused by the polychromatic spectra of X-ray tubes. In this method, two sets of scans, taken at different energy levels, were combined to create a synthetic monochromatic image. A physical polychromatic model of μ in dual energy imaging was developed with an iterative method to solve the model for a few selected pixels. To find a high-speed and effective computing approach, the physics model was approximated by a polynomial function of the measured intensities. The method was tested on a PMMA-aluminum phantom and CaCl₂ admixtures. The results show that streak and cupping artifacts are completely eliminated and that the measurement of the reconstructed attenuation coefficient μ is observed to be over 95% accurate.

Relevance feedback plays a key role in multiple feature-based image retrieval applications. This paper describes an online metric learning approach for a set of ranking functions. In the feedback round, the most relevant and most nonrelevant images related to the target image are selected to construct a relative comparison triplet. The weighting parameters of the multiple ranking functions are updated by minimizing a quadratic objective function constrained by the triplet. The approach unifies the learning algorithm for the most commonly used ranking functions. Thus, multiple features with their own ranking function can easily be employed in the ranking module without feature reconstruction. The method is computationally inexpensive and appropriate for large-scale e-commerce image retrieval applications. Customized ranking functions are well supported. Practically, simplified ranking functions yield better results when the number of query rounds is relatively small. Experiments with an image dataset from a real e-commerce platform show the superiority of the proposed approach.

Packet classification is crucial to the implementation of advanced network services that require the capability to distinguish traffic in different flows, such as access control in firewalls and protocol analysis in intrusion detection systems. This paper proposes a novel packet classification algorithm optimized for multi-core network processors. The proposed algorithm, AggreCuts, has an explicit worst-case search time with modest memory usage. The data structure of AggreCuts is flexible and well-adapted to different types of multi-core platforms. The algorithm on both Intel IXP2850 32-bit and Cavium OCTEON3860 64-bit multi-core platforms was implemented to evaluate the performance of AggreCuts. The experimental results show that AggreCuts outperforms the best-known existing algorithm in terms of memory usage and classification speed.