Can Basaran

Hop-by-hop congestion control and load balancing in wireless sensor networks

Due to the relatively high node density and source-to-sink communication pattern, wireless sensor networks (WSNs) are subject to congestion and packet losses. Further, the availability of low-cost hardware, such as Cyclops cameras, is promoting wireless multimedia sensing to support, for example, visual surveillance. As a result, congestion control is becoming more critical in WSNs. In this paper, we present a lightweight distributed congestion control method in WSNs. We develop new metrics to detect congestion in each node by considering the queue lengths and channel conditions observed in the one-hop neighborhood. Based on the estimated level of congestion, each node dynamically adapts its packet transmission rate and balance the load among the one-hop neighbors to avoid creating congestion and bottleneck nodes. In a simulation study performed in OMNeT++, our approach significantly enhances the end-to-end (e2e) packet delivery ratio and reduces the e2e delay without increasing the total energy consumption compared to the tested baseline approach.

Supporting Preemptive Task Executions and Memory Copies in GPGPUs

GPGPUs (General Purpose Graphic Processing Units) provide massive computational power. However, applying GPGPU technology to real-time computing is challenging due to the non-preemptive nature of GPGPUs. Especially, a job running in a GPGPU or a data copy between a GPGPU and CPU is non-preemptive. As a result, a high priority job arriving in the middle of a low priority job execution or memory copy suffers from priority inversion. To address the problem, we present a new lightweight approach to supporting preemptive memory copies and job executions in GPGPUs. Moreover, in our approach, a GPGPU job and memory copy between a GPGPU and the hosting CPU are run concurrently to enhance the responsiveness. To show the feasibility of our approach, we have implemented a prototype system for preemptive job executions and data copies in a GPGPU. The experimental results show that our approach can bound the response times in a reliable manner. In addition, the response time of our approach is significantly shorter than those of the unmodified GPGPU runtime system that supports no preemption and an advanced GPGPU model designed to support prioritization and performance isolation via preemptive data copies.

Grex: An efficient MapReduce framework for graphics processing units

In this paper, we present a new MapReduce framework, called Grex, designed to leverage general purpose graphics processing units (GPUs) for parallel data processing. Grex provides several new features. First, it supports a parallel split method to tokenize input data of variable sizes, such as words in e-books or URLs in web documents, in parallel using GPU threads. Second, Grex evenly distributes data to map/reduce tasks to avoid data partitioning skews. In addition, Grex provides a new memory management scheme to enhance the performance by exploiting the GPU memory hierarchy. Notably, all these capabilities are supported via careful system design without requiring any locks or atomic operations for thread synchronization. The experimental results show that our system is up to 12.4x and 4.1x faster than two state-of-the-art GPU-based MapReduce frameworks for the tested applications.

Quality of Service in Wireless Sensor Networks

Although well studied for traditional computer networks, quality of service (QoS) concepts have not been applied to wireless sensor networks (WSNs) until recently. QoS support is challenging due to severe energy and computational resource constrains of wireless sensors. Moreover, certain service properties such as the delay, reliability, network lifetime, and quality of data may conflict by nature. Multi-path routing, for example, can improve the reliability; however, it can increase the energy consumption and delay due to duplicate transmissions. Also, high resolution sensor readings incur more energy consumptions and delays. Modeling such relationships, measuring the provided quality, and providing means to control the balance is essential for QoS support. In this context, this chapter discusses existing approaches for QoS support in WSNs and suggests directions for further research.

Active queue management via event-driven feedback control

Active queue management (AQM) is investigated to avoid incipient congestion in gateways to complement congestion control run by the transport layer protocol such as the TCP. Most existing work on AQM can be categorized as (1) ad-hoc event-driven control and (2) time-driven feedback control approaches based on control theory. Ad hoc event-driven approaches for congestion control, such as RED (random early detection), lack a mathematical model. Thus, it is hard to analyze their dynamics and tune the parameters. Time-driven control theoretic approaches based on solid mathematical models have drawbacks too. As they sample the queue length and run AQM algorithm at every fixed time interval, they may not be adaptive enough to an abrupt load surge. Further, they can be executed unnecessarily often under light loads due to the time-driven nature. To seamlessly integrate the advantages of both event-driven and control-theoretic time-driven approaches, we present an event-driven feedback control approach based on formal control theory. As our approach is based on a mathematical model, its performance is more analyzable and predictable than ad hoc event-driven approaches are. Also, it is more reactive to dynamic load changes due to its event-driven nature. Our simulation results show that our event-driven controller effectively maintains the queue length around the specified set-point. It achieves shorter E2E (end-to-end) delays and smaller E2E delay fluctuations than several existing AQM approaches, which are ad hoc event-driven and based on time-driven control theory, while achieving almost the same E2E delays and E2E delay fluctuations as the two other advanced control theoretic AQM approaches. Further, our AQM algorithm is invoked much less frequently than the tested baselines.

Kintense: a robust, accurate, real-time and evolving system for detecting aggressive actions from streaming 3D skeleton data

Kintense is a robust, accurate, real-time, and evolving system for detecting aggressive actions such as hitting, kicking, pushing, and throwing from streaming 3D skeleton joint coordinates obtained from Kinect sensors. Kintense uses a combination of: (1) an array of supervised learners to recognize a predefined set of aggressive actions, (2) an unsupervised learner to discover new aggressive actions or refine existing actions, and (3) human feedback to reduce false alarms and to label potential aggressive actions. This paper describes the design and implementation of Kintense and provides empirical evidence that the system is 11% - 16% more accurate and 10% - 54% more robust to changes in distance, body orientation, speed, and person when compared to standard techniques such as dynamic time warping (DTW) and posture based gesture recognizers. We deploy Kintense in two multi-person households and demonstrate how it evolves to discover and learn unseen actions, achieves up to 90% accuracy, runs in real-time, and reduces false alarms with up to 13 times fewer user interactions than a typical system.

Energy-Efficient Privacy Protection for Smart Home Environments Using Behavioral Semantics

Research on smart environments saturated with ubiquitous computing devices is rapidly advancing while raising serious privacy issues. According to recent studies, privacy concerns significantly hinder widespread adoption of smart home technologies. Previous work has shown that it is possible to infer the activities of daily living within environments equipped with wireless sensors by monitoring radio fingerprints and traffic patterns. Since data encryption cannot prevent privacy invasions exploiting transmission pattern analysis and statistical inference, various methods based on fake data generation for concealing traffic patterns have been studied. In this paper, we describe an energy-efficient, light-weight, low-latency algorithm for creating dummy activities that are semantically similar to the observed phenomena. By using these cloaking activities, the amount of fake data transmissions can be flexibly controlled to support a trade-off between energy efficiency and privacy protection. According to the experiments using real data collected from a smart home environment, our proposed method can extend the lifetime of the network by more than 2× compared to the previous methods in the literature. Furthermore, the activity cloaking method supports low latency transmission of real data while also significantly reducing the accuracy of the wireless snooping attacks.

Kintense: A robust, accurate, real-time and evolving system for detecting aggressive actions from streaming 3D skeleton data

Kintense is a robust, accurate, real-time, and evolving system for detecting aggressive actions such as hitting, kicking, pushing, and throwing from streaming 3D skeleton joint coordinates obtained from Kinect sensors. Kintense uses a combination of: (1) an array of supervised learners to recognize a predefined set of aggressive actions, (2) an unsupervised learner to discover new aggressive actions or refine existing actions, and (3) human feedback to reduce false alarms and to label potential aggressive actions. This paper describes the design and implementation of Kintense and provides empirical evidence that the system is 11% – 16% more accurate and 10% – 54% more robust to changes in distance, body orientation, speed, and person when compared to standard techniques such as dynamic time warping (DTW) and posture based gesture recognizers. We deploy Kintense in two multi-person households and demonstrate how it evolves to discover and learn unseen actions, achieves up to 90% accuracy, runs in real-time, and reduces false alarms with up to 13 times fewer user interactions than a typical system.

Robust fuzzy CPU utilization control for dynamic workloads

In a number of real-time applications such as target tracking, precise workloads are unknown a priori but may dynamically vary, for example, based on the changing number of targets to track. It is important to manage the CPU utilization, via feedback control, to avoid severe overload or underutilization even in the presence of dynamic workloads. However, it is challenge to model a real-time system for feedback control, as computer systems cannot be modeled via physics laws. In this paper, we present a novel closed-loop approach for utilization control based on formal fuzzy logic control theory, which is very effective to support the desired performance in a nonlinear dynamic system without requiring a system model. We mathematically prove the stability of the fuzzy closed-loop system. Further, in a real-time kernel, we implement and evaluate our fuzzy logic utilization controller as well as two existing utilization controllers based on the linear and model predictive control theory for an extensive set of workloads. Our approach supports the specified average utilization set-point, while showing the best transient performance in terms of utilization control among the tested approaches.

Adaptive load shedding via fuzzy control in data stream management systems

Data stream management systems (DSMS) aim to process massive data streams in a timely fashion to support important applications, e.g., financial market analysis. However, DSMS can be overloaded due to large bursts in data stream arrivals and data-dependent query executions. To avoid overloads, we design a new load shedding scheme by applying distributed fuzzy logic control, which is very effective to deal with uncertainties in highly dynamic systems such as DSMS, based on the per-stream backlog and selectivity of each query operator. We have implemented our approach by extending an open source distributed DSMS. The performance evaluation using high-rate Internet traces shows that our approach closely supports a specified backlog bound for each data stream queue, while improving the query processing delay, with little overhead.