(A) scalable time slice CFS(Completely Fair Scheduler) scheduler through virtual runtime control for embedded system

김윤수 Graduate School of Convergence IT, Korea Universit 2011 국내석사

The CFS scheduler has been evaluated a good performance and stability scheduler in embedded system until now. The basic idea of CFS scheduler gives a fair time quantum to whole tasks. It means that the CFS scheduler allocates time quantum fairly according to each priority level; each task can occupy more time quantum through each priority. However, it is not suitable sometimes in embedded system. For example, a task A needs to a data or result value of low priority task B; task B is still running in system. A task A will wait the result of task B until finishing task B operation. If this phenomenon happen in embedded system or mobile phone, system occur the lock up or struggle state. My research focus can change the time slice CFS scheduler without changing a task priority. If /without changing priority at struggle state. This paper describes how the CFS scheduler controls each task’s time slice and shows the improvement result of elapsed time for phi calculation

Fault injection for testing device drivers using QEMU full system emulator

장익준 Graduate School of Convergence IT, Korea Universit 2013 국내석사

Device drivers are the most critical part of embedded system. Since device drivers should deal with asynchronous events from hardware, conventional testing methods are usually inappropriate to detect hidden defects in the complex execution path of device driver. This paper describes the low-cost method for testing device drivers using QEMU full system emulator. To introduce faulty conditions to device driver, faults are injected into virtual device. In addition, DMA operations of virtual devices, which are difficult to be monitored on real hardware, are checked against race condition between CPU and DMA operations for detecting DMA errors. Fault injector and memory tracing module were newly developed as QEMU extension. Experiments were carried out on the virtual ARM based embedded system to evaluate suggested test methods. Experiment results shows that it can readily simulates external faults for evaluating resilience of device drivers and detects race conditions between CPU and DMA master devices.

GPHTex : the GPHT with run-length monitoring

최재원 Graduate School of Convergence IT, Korea Universit 2013 국내석사

Up to this point, many researches had studied the DVFS (Dynamic Voltage and Frequency Scaling) to reduce power consumption of processors. The GPHT (Global Phase History Table) is a result of one such study that had presented an effective prediction mechanism for the utilization of the DVFS. However, the GPTH was not without a flaw. Because of its method to employ a fixed table for the prediction of the next DVFS option, the GPHT systematically cause two prediction misses when a same DVFS option repeats more than the window_size of the table. In order to solve this systematical problem, this paper introduces the GPHTex, the GPHT supplemented with run-length monitoring ability. This study has confirmed that the GPHTex indeed supplemented the GPHT with run-length monitoring ability without hindering the original prediction capability of the GPHT. This study, in its experiment, also confirmed that the GPHTex improves the prediction accuracy of the GPHT 8.98% at most and 3.28% at average. 현재까지 DVFS(Dynamic Voltage and Frequency Scaling)을 이용하여 프로세서의 전력소비량은 줄이기 위해 많은 연구가 진행되었다. DVFS를 효과적으로 사용하기 위해 최적의 DVFS설정의 예측 적중률을 향상시킨 GPHT (Global Phase History Table)또한 그런 연구의 결과물중 하나이다. 그러나 기존의 예측기법보다 상당히 높은 예측 적중률은 가지는 GPHT도 결점이 없지 않았다. 고정된 크기의 표를 만들고 그 표에 이전에 관측된 패턴과 패턴직후에 관측된 최적의 DVFS 값을 기록하여 이후의 예측에 사용하는 현재의 GPHT경우, 고정된 표의 Window_size보다 많은 수의 같은 DVFS값이 반복적으로 관측될 경우 최소 2번의 예측 miss를 만드는 구조적인 결함이 있다. 이와 같은 GPHT의 구조적 문제를 해결하기 위해 본 논문은 기존 GPHT에 Run-Length Monitoring 기능을 추가하여 GPHT를 보완한 GPHTex를 소개한다. 본 논문은 실험을 통하여 GPHTex가 GPHT 기존의 예측 적중률을 잃지 않으면서 성공적으로 GPHT의 문제점을 해결 하였다는 것을 확인하였다. 또한, 본 논문은 GPHTex가 기존 GPHT의 예측 적중률은 최대 8.98%, 평균 3.28% 향상시킨다는 것을 확인하였다.

Parsing method of JSON and protocol buffer at the Android platform

오범혁 Graduate School of Convergence IT, Korea Universit 2012 국내석사

As people who use the Internet through the smart phone increased, they started to complain about service speed. To fulfill such user requirement, several kinds of data-formats emerged such as XML. And to improve data transfer rate among servers which can satisfy users’ request and the device which the users use, another data-format called JSON was developed. Recently, Google developed a new data-format called Protocol Buffer. Our purpose was to offer users several kinds of knowledge about data-formats especially parsing method of three data-format and, through the experiments comparing between XML, JSON and Protocol Buffer, show that our parsing method was right or wrong and which one is the fastest data-format. To do this, we developed streaming server, parsing server and Android Application. By using these developments, we did experiments that compared three data-formats and observed that our development was useful when developers use our method and Protocol Buffer, above all other data-formats, is the fastest. Also at this paper, we explained the architecture of the new data-format and the method how can convert data-format and its’ example.

Vision algorithm for multi-core mobile devices

김은아 Graduate School of Convergence IT, Korea Universit 2012 국내석사

The recent emergence of multi-core processors and GPU has been enabled mobile device to employ various applications which are traditionally well known with heavy computation requirement such as vision application and mobile games. Computer vision (CV) applications such as video surveillance, camera-assisted intelligent driving are representative CV examples to take advantage of adopting high performance chipset for mobile platform as well. This report will explore that: (1) the characteristics of mobile applications; (2) introducing parallelization techniques that can be run improving performance both of multi-core CPU and GPU; (3); Eigenface recognition via accuracy improved PCA (Principle Component Analysis) [1]; (4) the summary of technologies that can be used in Tegra2[2]; (5) performance improvement by parallelizing matrix computation in Eigenface algorithm. The speed up by parallelizing face recognition software to use both of two cores in NVIDIA Tegra2 was overall 80% on Android [3].

Adaptive geographical routing in vehicular Ad Hoc networks

유정환 Graduate School of Convergence IT, Korea Universit 2013 국내석사

There has been considerable research for greedy geographical routing protocols which are promising in Vehicular Ad Hoc Networks where uses wireless communication technology. While greedy geographical routing protocols such as GPSR use fix hello message interval, 1.5 second, in VANETs, presented method is adaptive hello message which changes hello message interval to the number of neighbor table entry increases delivery rate up to 22%.

Position tracking for limited environment in GPS manner using pseudo satellites

류인호 Graduate School of Convergence IT, Korea Universit 2011 국내석사

(A) secure mobile payment protocol for wireless network

蔡秉鎬 Graduate School of Convergence IT, Korea Universit 2014 국내석사

TaintDroid extention : dynamic content provider tracking

金龍 Graduate School of Convergence IT, Korea Universit 2013 국내석사

(The) fast memory and File Management System (FMFS) for real-time applications

장진혁 Graduate School of Convergence IT, Korea Universit 2011 국내석사

In order to improve the performance of application, an optimization of algorithm, structure, and sequence typically used in application is used. However, such an optimization requires huge effort and time, and depends on the application, and therefore, a series of re-optimization should proceed in order to optimize other applications. In contrast, we introduce “The Fast Memory and File management System (FMFS) for real-time applications” that shows satisfactory performance in terms of memory and file processing without additional optimization of most of applications. As FMFS pre-allocates memory and file in sufficient size for an application before or immediately after the operation of an application and immediately performs service request of memory and file of an application on its run-time, it reduces overheads spent for the processing of memory and file allocation occurred on run-time and helps to improve the performance of an application.