Biker roadway safety plays a crucial role in the adoption of bicycling as a green form of transportation, and as a mode of active travel to promote increased physical activity for improved health and wellness. Encounters between cyclists and vehicles pose a particular threat to cyclists, whenever they are in close proximity to each other. In the CyberBike project, we propose to augment standard road bicycles with sensors and computational capabilities turning it into a cyber-physical bicycle. Towards this end, we are working to develop and evaluate a multi-modal sensing system (computer vision and audio processing) to detect vehicles in real time, as they approach a cyclist from the rear. This project is being executed by an interdisciplinary team of researchers, led by me, encompassing the areas of computer systems, computer vision, and exercise science.
The goal of this project is to investigate the feasibility of using Smart Phones as active elements in infrastructure-based mobile computing. Infrastructure-based mobile computing systems, such as Internet Suspend/Resume® (ISR) proffer a "carry-nothing" approach to personal computing, where a user is free to utilize any available computing device while still experiencing a fully personalized computing experience. Unfortunately, systems such as ISR and SoulPad (an infrastructure-free model) must choose between reliability, availability, and performance. Our approach is to combine the best qualities of these systems while eliminating their drawbacks. To accomplish this, we leverage a device people already carry wherever they go: their smart phones. We call such devices Horatio, named after Hamlet's trusted ally from the Shakespearean play. An Horatio device functions as the trusted ally for a user, improving the reliability (w.r.t. the SoulPad approach), and the availability and performance (w.r.t. the ISR approach) for their personal computing environments. Key contributions of this project are novel state transfer protocols, utilized to transfer user computing state between clients and Horatio, and between Horatio and servers. Additionally, we propose a number of optimizations to these protocols to improve interactive performance, to minimize smart phone battery usage during state transfers, and to reduce suspend and resume times.
Automating video-based surveillance requires addressing critical systems challenges in performance, scalability, and programmability, among others. Although computer vision algorithmic optimization may improve application performance and scalability, there are also potentially large gains in performance to be realized through the tailoring of resource management and scheduling to this domain. To address these challenges, we have proposed the Target Container, a system-level programming and execution abstraction that performs per target rather than per camera resource allocation for target detection and tracking in order to improve application performance and scalability. This project is a large multi-institutional collaboration between my advisor, Prof. Liviu Iftode, and a team from Georgia Tech. led by Prof. Kishore Ramachandran.
The goal of this project is to provide a framework upon which advanced tools can be built to manage modern network file systems. Our approach is to introduce a middlebox for network file systems (analogous to a network firewall or NAT) that interposes on the network path between network file system clients and servers, to enable the specification and enforcement of network file system access control and monitoring policies. Our research goals for this project include: (i) investigating and evaluating the feasibility of specifying and the performance of complex access control and monitoring policies executing within a network processing element, (ii) extracting models of user behavior through non-intrusive network file system access monitoring, (iii) designing and evaluating an expressive and powerful interface for specifying network file system access policies, which includes techniques for policy verification, and (iv) exploring techniques for graphically visualizing the outputs of complex network file system access policies.
A great number of people spend one or more hours each day driving between home and the office. These daily roadway commutes are highly predictable and regular, and provide a great opportunity to form virtual mobile communities. However, even though these commuters are already physically present in the same location, they are limited in their ability to communicate with each other. This project proposes a framework for building such communities, which we call Vehicular Social Networks (VSNs), to facilitate better communication between commuters driving on roadways. As a proof of concept, we have designed RoadSpeak, a VSN-based system that allows drivers to automatically join VSNs along popular roadways, and to communicate with each other by means of voice chat messages.
The goal of this project is to address the congestion problem in modern vehicular highways by applying concepts from the field of computer networking. Specifically, we are investigating the application of Quality of Service (QoS) techniques to introduce a reservation system for highways. We look to provide differentiated services by partitioning a portion of the highway (called a high-priority lane or HPL) and scheduling traffic on these HPL's.
The goal of this project is to investigate the performance of moving network file systems out of the kernel. This work is partly driven by recent advances in user-level networking technologies (e.g., Infiniband, Myrinet, etc.) Additionally, we are working on building a user-level NFS toolkit to aid in the further development of the popular NFS protocol and to aid in the evaluation of systems running NFS.
The goal of this project is to explore the use of non-conventional techniques in building systems that can perform failure detection and repair/recovery of state affected by a failure, while maintaining service to their clients. The cornerstone of this research is the use of Remote Memory Communication technologies (RMC). In contrast with previous research that has used RMC mostly for its performance benefits, we take a novel approach on using it as a building block in the design of remote healing systems.