Speaker: Mr. Seung-Jong Park , Georgia Institute of Technology

Title: Energy-Aware Protocols in Wireless Sensor Networks
Date and Time: April 29, 2004 2:30PM
Place: 164 Coates

Abstract:

Wireless sensor networks (WSN) typically consist of tiny sensors with a limited energy supply. This gives rise to the critical issue of energy conservation. This talk addresses the energy conservation issue by tackling three fundamental problems: network topology construction, robust downstream data delivery and optimal upstream data diffusion.

First, the talk will motivate the consideration of trade-offs between energy consumption and throughput in determining the optimal topology. Then a new topology control scheme, Adaptive Topology Control (ATC), which increases throughput per unit energy, will be presented.

Second, robust downstream data delivery is addressed while considering three unique challenges of WSNs: (i) Environment challenges: the limited lifetime of network nodes, the scarcity of bandwidth, energy and size of the network itself are characteristics different from those of wireless cellular or ad-hoc networks. (ii) Message size: most messages in a sensor network might be small sized queries which raise fundamental issues on what kind of loss recovery schemes should be employed. (iii) Reliability semantics: WSNs require other notions of reliability ranging from reliable delivery to only a sub-region of the network to partial probabilistic reliability for scoped-resolution based querying. The talk will address the above challenges and present an approach called GARUDA that provides reliable point-to-multipoint data delivery from the sink to the sensors. Simulation results will be used to prove how GARUDA can be a scalable framework with respect to the network size, message characteristics, loss rate, and reliability semantics.

Finally, the talk will consider a typical characteristic of WSN: correlation of data along the spatial, semantic, and temporal dimensions. Exploiting such correlation when performing data aggregation can result in considerable improvements in bandwidth and energy performance of WSNs. Two theoretical solutions, the shortest path tree and the minimum Steiner tree, will be introduced for the extreme cases of perfect and zero correlation. Then the talk will present a simple, scalable and distributed approach to construct a correlation-aware data aggregation structure in WSNs.

Biodata:

Seung-Jong Park is a Ph.D. candidate in the school of Electrical and Computer Engineering at Georgia Institute of Technology. His primary research interests are in the broad areas of wireless networks and mobile computing. He holds a B.S. from computer science of Korea University and a M.S. from computer science of KAIST (Korea Advanced Institute of Science and Technology). During his Ph.D. study, he has concentrated on the design of energy-aware protocols for wireless sensor networks. Before joining Georgia Institute of Technology, he had worked for Shinsegi Telecomm, which is the first CDMA cellular service provider in the world, as a research staff working on wireless 2G and 3G cellular networks.

Speaker: Mr. Mahesh Marina, State University of New York , Stony Brook

Title: Routing and Channel Assignment Algorithms for High-Performance Multihop Wireless Networks
Date and Time: April 27, 2004 3:30PM
Place: 155 Coates

Abstract:

In a multihop wireless network, every node acts as a router to enable communication between distant nodes that cannot communicate directly. The multihop feature distinguishes such a wireless network from traditional cellular networks and wireless LANs. We associate a broader meaning to the term "multihop wireless network" in the sense that it encompasses mobile ad hoc networks, wireless mesh networks and sensor networks. Multihop wireless networks are bandwidth-constrained as they are typically designed to operate on the limited unlicensed spectrum (Additionally, nodes may be energy-constrained; but we will not directly address the energy-efficiency issue here.). The effective bandwidth falls even further when we take into account other factors such as half-duplex radios and multiple access interference resulting from the shared nature of the wireless medium. Scarce bandwidth makes efficient utilization of available bandwidth through overhead reduction a driving factor when designing network protocols for routing and channel access. For the same reason, topology control techniques that improve network capacity by fully utilizing the given spectrum or through interference mitigation are equally important. This talk will consider routing in mobile ad hoc networks to emphasize the bandwidth-efficient protocol design aspect, and channel assignment in wireless mesh networks to highlight the capacity improvement via topology control.

In the first part of my talk, I will present an efficient, on-demand multipath routing protocol (termed AOMDV) for mobile ad hoc networks. Topology of a mobile ad hoc network can change rapidly and unpredictably because of arbitrary node movements. Routing protocols have to be simultaneously efficient as well as highly responsive to topology changes. On-demand protocols are a new class of protocols that reduce overheads by maintaining only needed routes a sharp contrast to the traditional proactive approach. AOMDV substantially improves upon the efficiency of basic on-demand routing by exploiting redundancy in network connectivity available in the form of multiple routing paths to efficiently cope with mobility-induced route failures. The novelty of AOMDV stems from its ability to find multiple disjoint and loop-free paths without using source routing or incurring high inter-nodal coordination overheads.

In the second part of my talk, I will address the problem of enabling a high-capacity wireless mesh network infrastructure. Simplistically, a wireless mesh network can be seen as a multihop extension of a standard wireless LAN, where every access point (AP) also acts as a router. While extending across multiple hops is good from the coverage viewpoint, it is bad in terms of capacity. Our approach is to improve capacity by using multiple non-interfering channels in conjunction with multiple radios per node. In fact, the IEEE 802.11 standard already has multiple channels available for simultaneous use. However, the medium access control (MAC) protocol in the standard is designed to operate over a single channel. Most of the previous work on enabling the use of multiple channels has focused on modifying the MAC while using a single radio interface; moreover, those solutions either require complex coordination tasks such as time synchronization or assume sophisticated radio reception capabilities. In contrast, our approach offers a more practical alternative by facilitating the use of multiple channels above the MAC layer and thereby allowing the use of commodity 802.11 hardware. I will introduce the channel assignment (radio-channel mapping) problem in this framework, present a novel formulation as a topology control problem, and describe an algorithm (termed MICA) to solve this problem. Towards the end, I will give a brief overview of my other research and outline my future research plans.

Biodata:

Mahesh Marina is currently a PhD candidate in Computer Science at Stony Brook University. He expects to graduate by August 2004. Previously, he received his MS (Computer Science) from the University of Texas at San Antonio in December 1999 and B.Tech (Computer Science & Engineering) from Regional Engineering College , Warangal , India in May 1998. His research interests broadly fall in the areas of computer networks and distributed systems, and specifically in the area of wireless networks. His interests span a wide variety of wireless networks including mobile ad hoc networks, wireless mesh networks, sensor networks and wireless LANs, and all layers of the network protocol stack above the physical layer.

Speaker: V.N. Venkatakrishnan, State University of New York, Stony Brook

Title: : Language-based enforcement mechanisms for system security
Date and Time: April 22, 2004 2:30PM
Place: 164 Coates

Abstract:

In this talk, I will discuss techniques that addresses threats to a host system from faulty or untrusted programs. In the first part of my talk, I will talk about an approach, that addresses the problem of protecting a host system's integrity from threats due to untrusted code using a technique that works by isolating the effects of the faulty/un-trusted program execution from the rest of the system In the second part, I will present a technique that addresses data and control flow related errors in a program (which often result in loss of data confidentiality and integrity). Here too, it is well known that this problem cannot be solved by use of execution monitors that employ pure runtime monitoring mechanisms. Hence, recent research has focused on pure static analysis (such as type analysis) for addressing this problem. The drawback of such approaches is that they tend to be overly conservative and result in a loss of precision, and reject many useful programs. In this work, we present an approach that uses a natural way to improve precision by augmenting static analysis with runtime checking. The use of this approach is illustrated with WinAgents (independently developed in Stony Brook), a framework for extraction and management of data over the Web. In addition, I will give a brief overview of my research background, and will discuss some ideas for future research work.

Biodata:

V.N. Venkatakrishnan is a PhD candidate in the Dept. of Computer Science, SUNY at Stony Brook. Venkat's research mainly focuses on computer security, with use of language based techniques and formal methods in a diverse set of problems in mobile code security, intrusion detection and protection of integrity and privacy of information. He received his M.S in Computer Science (2000) from Stony Brook University and a graduate degree in Mathematics (1997) from Birla Institute of Technology and Science (BITS), Pilani, India. His personal home page with links to publications and an informal biography is available at the foll. URL: http://www.seclab.cs.sunysb.edu/~venkat/

Speaker: Dr. Brygg Ullmer, Visualization Department, Zuse Institute Berlin

Title: Tangible tools for visualization, simulation, presentation, and collaboration
Date and Time: April 21, 2004 8:00AM
Place: 663 Life Sciences Annex

Abstract:

Scientific and information visualization have continued to gain ground as powerful means for understanding complex information. However, especially when combined with contexts such as virtual reality, teleconferencing, collaborative use, and grid computing, a mixture of anticipated and actual interface complexity often poses obstacles to use.

My doctoral and current research has developed an interface approach called "tangible interfaces." Tangible interfaces use systems of instrumented physical objects as a medium for interacting with digital information. I will briefly introduce some of my prior work, and focus on my current efforts toward using tangible interfaces to facilitate visualization, simulation, presentation, and collaboration. I will present examples from current applications in the computational sciences, including numerical relativity and surgical simulation. I will also consider how these tangible tools coexist with and complement parallel efforts toward web-mediated visualizations. I will conclude by discussing extensions to this work and future research directions.

Biodata:

Brygg Ullmer is pursuing a postdoctoral position involving visualization and grid computing at the Zuse Institute Berlin. He completed his Ph.D. at the MIT Media Laboratory in August 2002, where he studied with Prof. Hiroshi Ishii in the Tangible Media group. He also holds a B.S. in computer engineering from the University of Illinois, Urbana-Champaign (1994), and an M.S. from the MIT Media Laboratory (1997). He has been a visiting lecturer in Hong Kong Polytechnic University's School of Design (2002), and has held internships at Sony Computer Science Labs, Tokyo (2000) and Interval Research Corporation (1993-95). His research interests include tangible and graphical user interfaces for interaction with online media, complex infrastructure, biological systems, and group interaction contexts, as well as rapid physical and functional prototyping.

His homepage is visible at: http://www.zib.de/ullmer

Speaker: Gu-In Kwon, Boston University

Title: ROMA: Reliable Overlay Multicast with Loosely Coupled TCP Connections
Date and Time: April 20, 2004 3:30PM
Place: Coates 164

Abstract:

We consider the problem of architecting a reliable content delivery system across an overlay network using TCP connections as the transport primitive. For high-concurrency applications ranging from live streaming to reliable delivery of popular content, a recent research trend has proposed to serve these applications using end-system, or application-level, multicast. There is ample motivation for such an approach: multicast-based delivery provides excellent scalability in terms of bandwidth consumption and server load, while an end-system approach avoids the considerable deployment hurdles associated with providing multicast functionality at the network layer. This methodology has been successfully applied to develop best-effort, UDP-based methods for streaming applications, augmented with congestion control. At first glance, it seems that a similar approach can be applied to high-bandwidth applications requiring reliable delivery, merely by employing separate TCP connections at each application-level hop.

We first argue that natural designs based on store-and-forward principles that tightly couple TCP connections at intermediate end-systems impose fundamental performance limitations, such as dragging down all transfer rates in the system to the rate of the slowest receiver.

In contrast, the ROMA architecture we propose incorporates the use of loosely coupled TCP connections together with fast forward error correction techniques to deliver a scalable solution that better accommodates a set of heterogeneous receivers. The methods we develop establish chains of TCP connections, whose expected performance we analyze through equation-based methods. We validate our analytical findings and evaluate the performance of our ROMA architecture using a prototype implementation via extensive Internet experimentation across the PlanetLab distributed testbed.

Speaker: Dr. Gerald Baumgartner

Title: Compiler and Infrastructure Support for High-Performance Computing: The TCE and the Organic Grid
Date and Time: April 15, 2004 3:30PM
Place: Coates 155

Abstract:

The performance of large scientific computations is influenced by a variety of factors: from compiler optimizations to library design to the scheduling strategy on (collections of) machines. In this presentation, we will examine two very different approaches to using the available computing resources effectively: the data locality optimizations in the Tensor Contraction Engine (TCE) and the task scheduling on the Organic Grid.

The accurate modeling of the electronic structure of atoms and molecules in quantum chemistry involves computationally intensive tensor contractions over large multi-dimensional arrays. We are developing a compiler, the TCE, that translates a high-level specification of such computations into efficient, parallel code tailored to the characteristics of the target architecture. One of the optimizers in the TCE is concerned with reducing the disk I/O cost for accessing large intermediate arrays using a combination of loop fusion and loop tiling. We present a cost-model driven optimization framework for searching among the space of fusion and tiling choices. The effectiveness of the approach is demonstrated using examples from quantum chemistry.

Desktop grids have been used to perform some of the world's largest computations, but they have been centralized and tailored to an individual application. We are developing a fully decentralized approach to the organization of computation that is based on the autonomous scheduling of strongly mobile agents on a peer-to-peer network. Contrary to the popular master/worker organization of current desktop grids, our approach does not rely on specialized super-servers or on application-specific clients. By encapsulating computation and scheduling behavior into mobile agents, we decouple both application code and scheduling functionality from the underlying infrastructure. The mobile agents act as carriers that deliver the application to machines with available resources using (potentially application-specific) scheduling code. The resulting system is one where every node can start a large grid job, and where the computation naturally organizes itself around available resources. We have demonstrated the feasibility of this approach using the BLAST genome sequence alignment application and using a Cannon-style matrix multiplication.

Currently, these approaches are specialized to tensor contractions and desktop grids, respectively. We are planning to generalize them to optimizing arbitrary array computations and to discovering resources on arbitrary networks, respectively.

Speaker: Minaxi Gupta, Georgia Institute of Technology

Title: As You Give, So Shall You Receive: Reputation Driven Service Differentiation in Peer-to-Peer Networks
Date and Time: April 13, 2004 2:30PM
Place: Coates 164

Abstract:

Peer-to-Peer networks like Gnutella and Kazaa have revolutionized the concept of content distribution. In contrast with the traditional client-server paradigm, each participant in these networks is capable of being a server in addition to being a client. The result is that these networks have opened up new possibilities for resource sharing in the form of content, storage space, and processing capacity. However, several challenges stand in the way of effectively harnessing the power of peer-to-peer technology. In particular, the lack of centralized administration makes it hard to distinguish cooperating participants from "free-riders". Further, no inherent mechanisms exist in these networks to motivate participants to contribute to the common good of the system. In this talk we consider how to address these problems by augmenting peer-to-peer systems with service differentiation mechanisms. These mechanisms allow peer behavior to be tracked and insure that the quality of service (QoS) a peer receives is a function of its own behavior. Our approach integrates three components in its design. First, I will describe a set of QoS parameters that are relevant in the peer-to-peer context. I will then discuss the design of a reputation system that reliably tracks participant contributions. Finally, I will present a protocol to accomplish service differentiation that utilizes peer reputations as a substrate. Simulations show that the overheads in reputation computations are controllable because highly accurate reputations are not necessary. They also show that service differentiation is achievable using the proposed mechanisms.

Biodata:

Minaxi Gupta is a PhD candidate in the College of Computing at Georgia Institute of Technology. Her primary research interests are in the broad areas of networking and distributed systems. She holds a B.Sc. from Bombay University and an M.Sc. form IIT Bombay, both in Physics, and an MS in Computer Science from Georgia Institute of Technology.

Speaker: Dr. J. Edward Swan II

Title: Functional X-Ray Vision and Active Augmented Reality Systems
Date and Time: April 12, 2004 3:30PM
Place: Coates 152

Abstract:

At the Naval Research Laboratory, we have spent the past 6 years developing a mobile, outdoor augmented reality (AR) system, termed BARS, for heads-up situational awareness in urban settings. AR systems render heads-up information which is superimposed upon, and correlated with, a view of the real world. Having built this system, a major scientific objective is understanding how people perceive and process AR information, and how they might best use this information to perform tasks of interest to the US military. In this talk I will first briefly describe our system, and then I shall describe two recent empirical experiments, conducted at the Naval Research Laboratory and at Virginia Tech.

The first experiment investigates Function X-Ray Vision. A major phenomena that arises from AR, which does not appear with any other computerized display technique, is the ability to see information occluded by solid structures; for example, the routing of wiring behind a solid wall. Although Superman has possessed x-ray vision since the 1930s, augmented reality allows anyone to have the functional equivalent of this ability. Given that we can draw potentially many layers of occluded structure behind a solid surface, the research question is how should these layers be rendered so that a user can clearly perceive the resulting three-dimensional structure? In this experiment we propose and evaluate several such techniques.

The second experiment investigates an aspect of how we can use AR systems in actual outdoor settings. In particular, two challenges of outdoor environments are widely varying illuminance, and widely varying background textures. We (and others) have proposed that it is necessary for an outdoor AR system to sense the user's environment, and consider this information when determining how to render the system's graphics. We have termed this an "Active AR" system. We have performed an experiment where we modified the color of augmented text according to the color of different real-world backgrounds. Our experiment gives the first empirical evidence of how an Active AR system should render its graphics according to changing background conditions.

Biodata:

Dr. Swan is a scientist with the Naval Research Laboratory, where he is affiliated with the Virtual Reality Lab. His research has been broad-based, centering on the topics of virtual and augmented reality, computer graphics, empirical methods, visualization, human-computer interaction, and human factors. Dr. Swan has lead research efforts in information presentation and interaction techniques, including the perception of occluded objects in augmented reality, navigation and manipulation in virtual reality, the perception of color and texture in desktop displays, and an interface for visualizing brain and cranial base tumors. He has also been active in formalizing usability engineering methodologies and evaluation techniques for augmented and virtual reality. In computer graphics, he has investigated algorithms for efficient and accurate rendering of volumetric datasets, rendering architectural datasets for virtual reality applications, and image-based rendering techniques for efficient terrain visualization. Dr. Swan's current work is motivated by the problem domain of worn and desktop systems for command and control and battlefield visualization; previous motivating problem domains include biomedical visualization, terrain rendering, and virtual reality. Dr. Swan has a Ph.D. from Ohio State University, where he studied computer graphics and human-computer interaction. Dr. Swan has been active in the IEEE Visualization conference; he served on the conference committee for eight years, including Program Co-Chair in 2001 and 2002. Dr. Swan is a member of ACM, SIGGRAPH, SIGCHI, IEEE, and the IEEE Computer Society.

Speaker: Dr. Subodh Kumar, Johns Hopkins University,Department of Computer Science

Title: SCALABLE WALKTHROUGH
Date and Time: April 2, 2004 3:30pm
Place: 145 Coates

Abstract:

The need to navigate through increasingly complex virtual environments for simulation based design, and its validation, has motivated research in computer graphics for long. While hardware performance has grown at an impressive rate, model sizes have grown faster. In order to achieve realistic interactive walkthroughs we need to take a comprehensive approach and enrich models with rendering enhancing information.

In this talk I will present an algorithm to render increasingly large spline and polygonal models. I will describe parallel pre-processing methods to generate data useful for efficient rendering algorithms. I will discuss multi-resolution model representations, dynamic tessellation of parametric surfaces and polygonal simplification. I will present both point and region based visibility algorithms and introduce the notion of vLOD: the visible level of detail. Finally, I will argue that efficient pre-processing and data compression methods enable scalable output sensitive display algorithms that that do not necessarily slow down with increasing model sizes.

Speaker: Dr. Subodh Kumar, Johns Hopkins University,Dr. Guofei Jiang, Dartmouth College

Title: Dynamic Integration of Distributed Semantic Services in Network-based Computing
Date and Time: April 1, 2004 3:30pm
Place: 155 Coates

Abstract:

Global networking is changing the way that we think about and perform computation. Grid computing may link tens or hundreds of distributed devices, sensors and computing resources to support an application cooperatively. One critical challenge is how to dynamically discover and integrate these distributed resources for various applications. In recent years many distributed computing technologies have been emerging to support such large-scale network-based computing. To improve semantic interoperability among distributed systems, markup languages such as eXtensible Markup Language (XML) and Web Ontology Language (OWL) are used to create ontologies and describe data semantics. Loosely coupled distributed computing technologies such as web services enable distributed systems to be dynamically interfaced in a networked environment. Employing these cutting-edge technologies, we developed a new framework for dynamic integration of distributed services in network-based computing. Semantic Web technology is used to describe, locate and compose distributed semantic services across network. Distributed computing and data resources are organized with a peer-to-peer overlay network for scalability and reliability. Web services and content-based routing are employed to integrate the distributed services "on the fly".

This talk will introduce our prototype system and also cover some of our theoretical research on distributed computing: Semantic interoperability and information fluidity of large-scale systems; Performance modeling and comparison between data movement and code movement in distributed computing; Functional validation of distributed services in computational grid etc..

Speaker: Dr. Yong Liu, University of Massachusetts

Title: Large Scale Communication Networks: Modeling, Control and Optimization
Date and Time: March 31, 2004 3:30pm
Place: 155 Coates

Abstract:

Over the past 15 years, communication networks, and the Internet in particular, have grown exponentially in many dimensions including size, speed, heterogeneity and application-use. The analysis and management of such complex systems is of critical importance, but presents many challenges. In this talk I will present my work on modeling, control and optimization of large scale communication networks.

The first part of this talk is devoted to scalable modeling and simulation of large, high-bandwidth IP networks. I will present abstract dynamical models of network components, including the Transport Control Protocol (TCP) at the edge and Active Queuing Management (AQM) schemes in the core. Dynamical models enable us to obtain key performance metrics, such as queuing delay, packet loss probability, of large high bandwidth networks. Subsequently, I will present the design, implementation and performance of our model-based fluid simulations and hybrid fluid/packet simulations for large IP networks. These fluid models also pave the way for systematic analysis and design of network congestion control schemes.

The second half of my talk addresses the problem of interacting layer of control (network-layer and application-layer) in routing in overlay networks. Overlays are increasingly being used to deploy network services that cannot be easily embedded in the underlying Internet. Our study shows that iterative selfish route optimization in the overlay combined with network-wide route optimization in the network layer can result in oscillations and degrade overall network performance. We then propose strategies for overlay networks that eliminate oscillations but allow overlays to optimize their own performance.

My talk will be concluded with an overview of other projects I have worked on and directions for my future research.