|
|||||||||||||||
 |
EventsGraduate SeminarSpring 2004 ScheduleLocation: LOW 4034
AbstractsScott Coull - Trust in Routed Networks Abstract: The Internet has grown to be a very dangerous place. Internet worms cause heavy amounts of traffic while infecting new hosts, all while your inbox fills with junk e-mail. Reputation-based trust systems have become the cornerstone of most e-commerce sites, including eBay.com. In this talk, I describe a gossip-based system to protect not only your local network, but also the surrounding network infrastructure from the effects of various types of malicious network traffic using a reputation-based trust model. Anil Karanam - Software framework for efficient implementation of variable p in stabilized FEM Abstract: Recently, a stabilized finite element solver has been developed to exploit the hierarchical basis capability to grade polynomial order while maintaining C0 continuity. The hierarchical basis accomplishes this by starting with vertex interpolants (a linear basis) and then allowing the polynomial order to vary on each entity (edges, faces, and regions) in the mesh which are then multiplied by blends within each element to build a composite function that is locally higher order but still globally continuous. Details of this formulation and its efficient implementation will be presented. A preprocessing based framework is presented which deals with efficiency issues include maintaining the templated structure by grouping elements based on the maximum polynomial order of its entities. This model gives us the ability of these solvers to exploit the latest developments in model and mesh representations without sacrificing efficiency or having to extensively rewrite of the core solver. Matt Schumaker - Matrix Visualization of Graphs Abstract: In recent years both industry and academia have started to produce massive amounts of data. Computers have played a key a role in both the production and storage of this data. Much of this data comes in the form of relational information that codifies connections between objects. The data may hold relationships about genes, consumer shopping habits, economic indicators or terrorists. The traditional means of storing these relationships is in a graph. The knowledge that can be obtained from these graphs carries both academic and economic value. However, displaying these graphs can be troublesome as more and more information is added. Many techniques in use today do not do an adequate job of maximizing what is presented to the user. For this reason we investigated a novel visualization technique called "Matrix Visualization". This technique presents graph information in a very dense manner. To perform this investigation we developed a software architecture to create the visualization then performed a validation study to compare it to traditional graph visualization techniques for three common tasks: cluster identification, tree level identification and intra-level relationship identification. After doing the study, we found that the matrix visualization outperforms traditional visualization techniques in terms of speed or accuracy for these tasks, in some cases as much as a factor of two. This work contributes a foundation of new understanding of this little researched technique as well as adds to the existing literature new knowledge of traditional graph visualization. Vinay Nadimpally - Graph Theoretic Models for Protein Folding Abstract: A structured folding pathway, which is a time ordered sequence of folding events, plays an important role in the protein folding process and hence, in the conformational search. Pathway prediction, thus gives more insight into the folding process and is a valuable guiding tool to search the conformation space. We propose a novel "unfolding" approach to predict the folding pathway. We apply graph based methods on a weighted secondary structure graph of a protein to predict the sequence of unfolding events. When viewed in reverse this yields the folding pathway. We demonstrate the success of our approach on several proteins whose pathway is experimentally known. Garrett Yaun - Effecient Large-Scale Computer Systems and Network Models Using Optimistic Parallel Simulation later Abstract: We demonstrate our approach for enabling allowing large-scale simulation models to achieve greater scalability and performance. Our major contribution was obtained from the examination of large-scale optimistic simulation models and led to a reduction in both space and execution time as well as greater scalability. The models consisted of network protocols and distributed computer system applications across the Internet. The applications focused on were file transfers and a configurable application view storage system (CAVES). The protocols examined were TCP and IP. Within these models, a new technique called reverse computation was an important component in achieving performance gains and dispelling views that optimistic techniques operate outside of the performance envelope for Internet protocols. These are the first real-world models to leverage reverse computation and demonstrate its efficiency. Our experiments show that these models perform well on top of the an optimistic simulation engine called Rensselaer's Optimistic Simulation System (ROSS). We propose the construction of Rensselaer's Optimistic Simulation System Distributed (ROSSD). ROSS is designed for shared memory parallel computers and thus is limited by the number of processors. The most resent supercomputers are constructed on distributed computer technology, i.e. a super-cluster. To date, efficient optimistic performance has not been shown on large super-cluster platforms. In order to achieve even greater scalability and more accurate models of the Internet, a distributed version of ROSS is essential. ROSSD will be implemented using the Libsynk library from Georgia Institute of Technology. This library will provide the functionality for the handling of remote communication and Virtual Time synchronization. The use of graph partitioning algorithms will be used for mapping of simulation entities a cross the distributed system. With ROSSD, we hypothesize the simulation models can execute efficiently on some of the world largest distributed computers. In addition we propose the implementation and testing a reverse computation memory library. This library will allow for easier implementation of simulation models, allow for the models to use dynamic memory and allow for an over all reduction of memory in the simulation models as compare to models that work statically ``pre-allocated'' call necessary memory. |
|
|||||||||||||
News
