Current Appointment: Professor: Dept of Computer Science, Kent State University, Now

Major Research Project Accomplishments:

Fast TCP: FAST TCP is a variant of TCP networking protocol, developed at the California Institute of Technology, with a congestion control algorithm especially targeted at high-speed, long-distance links. It won the 2004 Internet speed record award given by Internet2. To demonstrate the competitiveness and flexibility of transparent and interactive networking, we have re-implemented equivalent of the FAST TCP over TCP interactive at application layer using transientware. The advantage of the system is that it is highly extensible and does not require kernel modification-- yet it matched the performance of FAST TCP. The approach, instead of requiring custom reactive codes modification inside TCP kernel used only event based application level interactive access. A FreeBSD Linux implementation of the proposed interactive OSI kernel including an extensible transientware mechanism has been realized.

Interactive Transparent Networking Model: We have proposed a new cross-layer networking model which extends classical OSI network kernel organization by enabling event based cross-layer interaction and an active extension mechanism called transientware. The advantage of the system is that it is extensible-- yet much less radical approach than active networking. The approach, instead of allowing custom reactive codes to be placed inside network or system layer- proposes only event based application level interactive access into the protocol state information. The actual ‘custom actions’ are then performed by special programmable components called transientware at application space. It relives lower network layers from housing costly custom components and addresses other practical issues like security and flexibility. A FreeBSD Linux implementation of the proposed interactive OSI kernel including an extensible transientware mechanism has been realized. Dramatic cross-layer optimizations have been demonstrated using a number of TCP/IP family protocols with live internet experiments. Sponsored by DARPA.

Universal Extensible Socket: Led the development of a composable socket system, where the socket layer at the communicating hosts can be extended to implement various advanced network transport systems. Formally, it demonstrated a modular construction formalism where complex transport constructs can be built using lower level isomorphic constructs. The demonstrated transports include store-and-forward transport (uses multiple TCPs to create a high level TCP lookalike which can handle data transfer on periodic intermittent connection) and concurrent communication transport- (uses multiple UDP/TCPs to set up multiple virtual paths to obtain high QoS communication on a network graph). Both systems have been demonstrated.

Interactive Infrastructure Less Rapid Handoff For Next Generation Mobility: TCP and other connection oriented transports used by applications suffer major performance drawbacks when they have to go through reconnection while in mobility. Mobile IP is logically capable of offering disconnection free handoff- but in practice the long delay causes severe ill-reaction with network’s congestion control mechanisms. In this work we have taken FreeBSD networking kernel with Interactive Transparent Networking Model and demonstrate a new mobility mechanism that offers fast seamless reconnections based on proactive event triggered actions. The scheme- Interactive Protocol for Mobile Networks (IPMN) uses rapid cross-layer interactivity. It probes the link layer and intelligently reconfigures the address. The connection is renewed to handle this address change by manipulating the TCP/IP stack at the end-points. It also does not need any infrastructure. Both with real internet experiment and detail model based performance comparison with mobile IP we have shown the IPMN dramatically outperforms Mobile IP.

Semantic Course Composer: Under NSF MatDL project, we have recently demonstrated a semantic composition engine called “Course Composer” for Material Science Collection. “Composition” such as course planning at the moment is considered a sophisticated cognitive task and requires human. We have recently developed a software system that can analyze concept dependencies and based on that help teachers to develop various dynamic teaching materials for structured presentation of thematic concepts from collections. Given a time constraint, desired level of depth in a thematic area, student background, the system analyzes the semantic concept among various sub-topics dependencies, and the format and pedagogical constraints, and suggests an automatic composed material collections that can provide high level help to design various courseware for rapid and targeted dissemination of knowledge. This concept system can create various pedagogical teaching styles such as top-down, bottom-up, iterative reinforcement presentations. A prototype has been presented the NSF Digital Library Annual Conference’s CritLab 2004.

Polymorphic Data Fusion & Visualization: Under the current ITR grant the PI has developed a prototype algorithm and associated system for visualization of polymer material from multiple instrument perspective. Each polymer observation instrument is bound by various physical and experimental constraints. The system aims at developing a generalizable data fusion framework where data from multi-instrument based techniques can be synthesized to eliminate the ‘blind spot’ of any individual technique, and to provide a complete view. The prototype now allows automatic 3D model construction from information fused from X-ray Photoelectron Spectroscopy (XPS), and 3D Laser Scanning Confocal Microscopy (LSCM) fusion. Shortly it will allow fusion of Atomic Force Microscopy (AFM) data offering researchers unprecedented integrative view into polymers.

Perceptual Transcoding for Visual Teleporting: This ‘science fiction’ like system uses live human perceptual foveation (eye gaze information from a real time eye-tracker) and an MPEG-2 network transcoder utilizes such fine precision information for precise bandwidth reduction/adaptation for serving broadcast quality video over local narrow links. The perceptual transcoder achieves real-time transcoding speed by using latest MV-bypass techniques and multiple off-the-shelve high end PCs. One of the first system to incorporate live eye-gaze based transcoding for video. Sponsored by DARPA, Demonstrated in ACM Multimedia 2001.

Self-Morphing Active Video Steaming: A socket like construct which can be used by applications like a conventional socket- but it has embedded sensor intelligence and active computing capability- it can transform video while carrying it- based on the characteristics of various links through which video travels. Active Video Stream (AVIS) sockets- armed with a full logic parallel MPEG-2 encoder implemented in it- acts like a reusable transport but can perform arbitrary video transformation while carrying MPEG-2 video content over a quasi-active network. The video stream may gets rate transformed through distribute active computing even while the packets diffuses over an active subnet via multiple paths. In a way it creates the illusion of an self-morphing video stream. It can operate over a quasi-active network with only a few sparsely ‘floating’ active junction nodes in a ‘sea’ of conventional switches and junctions. The work has been designated as the concept application for the “High Performance Active Networking Team” within the DARPA active networking initiative (ANI). The prototype-- splicing a wireless network with high-bandwidth vBNS has been demonstrated in the DARPA Active Network Conference and Exposition, May, 2002.