Adam Taylor will present and defend his Masters Project Defense Friday, November 20 at 1:00 p.m. in Avery 347. The title of his defense is "On Problematic Robotic Thresholds."
Abstract
With the explosive growth of social networks, vast amounts of data are being generated every second. Twitter plays an important role among the social networks through huge information flow among its millions of users. Geographic information encoded in the data can be useful in many applications including pre-election polling, studying linguistic patterns, determining sickness trends, responding to natural disasters, and sentiment analysis in specific regions. However, only a small percentage of users expose their geographic coordinates.
In this project, we have implemented an algorithm to estimate the geolocation of twitter users, independent of their individual location sharing preferences using publicly available Twitter data. We have also extensively analyzed the performance of the algorithm using three different types of spatial clusters, and different connectivity patterns. Results show that this approach can accurately estimate the geolocation of active users in sufficiently dense networks.
---------------------------------------
Guangdong Liu will present and defend his Masters Defense Project on Monday, November 23 at 2:00 p.m. in 211 Schorr Center. The title of this defense is "An Efficient Fault Recovery Algorithm in Multiprocessor Mixed-Criticality Systems."
Abstract
Recent years, there is an increasing interest of integrating mixed-criticality functionalities onto a shared computing platform in automotive, avionics and the control industry.The benefits of such an integration include reduced hardware cost and high computational performance. Also, new challenges appear as a result of the integration since interferences across tasks with different criticalities are introduced and these interferences could potentially lead to catastrophic results. Failures are likely to be more frequent due to the interferences. Hence, it is becoming increasingly important to deal with faults in mixed-criticality systems. Although several approaches have been proposed to handle failures in mixed-criticality systems, they come either with a high cost due to a hardware replication (spatial redundancy) or with a poor utilization due to re-execution (time redundancy).
In this project, we study a scheme that provides fault recovery through task reallocations in response to permanent faults in multiprocessor mixed-criticality systems. We present an algorithm to minimize the number of task reallocations while retaining the promise that the most critical applications continue to meet their deadlines. The performance evaluation of the proposed algorithm is carried out by comparing it with two baseline algorithms. In order to evaluate the performance of algorithms from the perspective of mixed-criticality systems, we choose the state of art metric called ductility to formally measure the effects of deadline misses for tasks with different criticality levels. Under this metric, a high-criticality task is considered more import than all low-criticality tasks combined. The simulation results confirm the effectiveness our proposed algorithm in both minimizing the number of task reallocations and retaining the promised performance of high-criticality tasks.
--------------------------------------
Adrian Lara will present and defend his Masters Defense Project on Tuesday, November 24 at 9:30 a.m. in 211 Schorr Center. The title of this defense is "Using Software-defined Networking to Improve Campus, Transport and Future Internet Architectures."
Abstract:
Software-defined Networking (SDN) promises to redefine the future of networking. Indeed, SDN-based networks have unique capabilities such as centralized control, flow abstraction, dynamic updating of forwarding rules and software-based traffic analysis. SDN-based networks decouple the dataplane from the control plane, migrating the latter to a software controller. By adding a software layer between network devices and applications, features such as network virtualization and automated management are simpler to achieve.
In this dissertation, we show how SDN-based deployments simplify network management at multiple scales such as campus and transport networks, as well as future Internet architectures. First, we propose OpenSec, an SDN-based security framework that allows network operators to implement security policies in campus networks. Second, we propose the eXtensible Traffic Engineering Framework (XTEF) to enable application-driven traffic engineering and provision transport network resources using on-demand Wavelength Division Multiplexing (WDM) tunnels. Third, we demonstrate how SDN can be used to dynamically create intra-domain cut-through switching tunnels to bypass the routing layer in MobilityFirst. Finally, we propose how to extend the cut-through capabilities to inter-domain routing in MobilityFirst.
In our work, we run experiments on the GENI testbed (Global Environment for Network Innovations), the ORBIT (Open-Access Research Testbed for Next-Generation Wireless Networks) and Mininet. The results show that SDN can be used to simplify policy-based network management, virtualize an entire WAN as a single switch, create Wavelength Division Multiplexing (WDM) tunnels on demand and create inter-domain tunnels using techniques that scale better than traditional distributed methods.
--------------------------------------
Michael Rogers will present and defend his Masters Defense Project on Monday, November 23rd at 11:30 a.m. in 347 Avery. The title is "Transforming C OpenMP Programs for Verification in CIVL".
Abstract:
There are numerous ways to express parallelism which can make it challenging for developers to verify these programs. Many tools only target a single dialect but the Concurrency Intermediate Verification Language (CIVL) targets MPI, Pthreads, and CUDA. CIVL provides a general concurrency model that can represent pro- grams in a variety of concurrency dialects. CIVL includes a front-end that support all of the dialects mentioned above. The back-end is a verifier that uses model checking and symbolic execution to check standard properties.
In this thesis, we have designed and implemented a transformer that will take C OpenMP programs and transform them to CIVL so that they can be verified. A large subset of the OpenMP specification is supported by the transformer. The transformer operates on a Abstract Syntax Tree (AST) representation of the program. The transformer will modify the AST to replace OpenMP constructs with equivalent CIVL constructs so that the program can be verified in CIVL. Results show that the transformer supports the most common used OpenMP constructs.