Bongen Gu

The Design and Implementation of a GPS-based Realtime Unmanned Driving System

This study is to design and implement an unmanned driving system that runs to a certain destination of the user in connection with a smart phone and Arduino. Once the departure and destination positions of the unmanned vehicle are set by means of a smart phone, it runs based on GPS information. This system is designed in a way that the monitoring and controlling are also possible by means of a smart phone. An Arduino-based unmanned vehicle is also designed and implemented

Communication and Computation Overlap through Task Synchronization in Multi-locale Chapel Environment

Parallel processing systems use data parallelism to achieve high performance data processing. Data parallelism is normally based on data arrays, which are distributed to separate nodes. Therefore, efficient communication between nodes is required to initialize the distribution. In this paper, we propose a computation and communication overlapping technique to reduce the overhead of communication during array distribution. Our overlapping technique uses task parallelism for the initiate task and the worker tasks, and also the synchronization mechanism supported by Chapel. To show our overlapping technique is effective, we design and develop a parallel version of the Mandelbrot set program, and then evaluate the benefit of overlapping against the execution time of Mandelbrot. From our comparison, the overlapping technique proposed in this paper is effective in that it reduces the impact of communication in the initial array distribution.

Potentiality for Executing Hadoop Map Tasks on GPGPU via JNI

Hadoop has good features for storing data, task distribution, and locality-aware scheduler. These features make Hadoop suitable to handle Big data. And GPGPU has the powerful computation performance comparable to supercomputer. Hadoop tasks running on GPGPU will enhance the throughput and performance dramatically. However the interaction way between Hadoop and GPGPU is required. In this paper, we use JNI to interact between them, and write the experimental Hadoop program with JNI. From the experimental results, we show the potentiality GPGPU-enabled Hadoop via JNI.

GPUedHadoop : Hadoop as Parallel Processing Framework for GPU Cluster

Many research groups try to use GPGPU to enhance the performance of Hadoop. In this paper, we propose new approach to enhance the performance of Hadoop Map task and Combiner by using GPGPU on Hadoop Cluster. Our approach is that the whole HDFS block called split is passed to Map task for GPU processing. And then, the result of Mapper enabling GPU processing is also passed to Combiner for GPU processing. In other words, accelerated steps via GPU are Mapper and Combiner in Hadoop. GPU-enabled Hadoop adopting our approach has the same characteristics as native Hadoop, and additionally high performance feature. To show that our approach is effective to enhance the performance of Hadoop by using GPU, we experiment on GPU-accelerated Hadoop. Our experimental results show that speedup factor of our approach is between 3.27 and 4.19. So, we can conclude that our approach for GPU-enabled Hadoop is effective to enhance the performance.

MAP task allocation strategy in an ARM-based Hadoop cluster by using local storage as split cache

The increase of power consumption makes the cost of cluster operation higher. One approach for reducing power consumption is to establish a cluster with small nodes which equip a low-power, high-performance processor. Since many low-power consumed nodes do not have storage devices, a separate storage system is required to store large-volume data while nodes mount this storage space to save data. When a Hadoop cluster is configured in such a condition, each nodes access to a storage results in excessive network load and delays the execution of Hadoop Map tasks. In this study, we propose a newmap task scheduling policy for Hadoop. This policy transmits multiple splits to nodes at once to reduce network load. In addition, local storage space of nodes is used as a cache for a split, which shortens the time to access splits, so this policy can reduce the execution time of Hadoop applications.

Configurations of Dual RAID System

The RAID system is used to get the high performance and reliability of disk system. The many RAID system concerned the disk failure, and have the recover policy. But the probability of the RAID system failure is not lower than that of the disk failure. To implement the robust RAID system, the redundancy of RAID controller is also needed. In this paper, we provide three configurations of Dual RAID system. And we consider the characteristics of each Dual RAID system.

Performance Analysis of RAID Implementations

There are many RAID configurations and two types of RAID implementations. To decide which configuration or implementation is suitable to an application, the performance information of RAID system are required. In this paper, we evaluate the performance of RAID system. For evaluating the performance, we use many parameters composed of two RAID configurations, two implementations, and two types of storage devices. Two configurations, which we use, are level 0 and 5. Two implementations are hardware and software RAID. As a result, if we want to construct the RAID level 0, it is good decision to use the software RAID. And the overhead of the software RAID level 5 is higher than that of hardware RAID.