| 題目：SSBDS – A High-Performance Big Data Service for Research and Education 主講人：鍾葉青 教授 (國立清華大學資訊工程學系) 時間：104年3月24日(星期三13：30 - 15：00) 地點：文1F11 Abstract ： UniCloud is a distributed cloud system, which consists of a number of cloud platforms at different universities, for research and education in Taiwan. In the UniCloud system, several cloud platforms can form a community cloud. Also, a cloud platform in the UniCloud system can leverage the pubic cloud resources to form a hybrid cloud. The services provided by the UniCloud system include SSCloud – a VM service, SSBox – a cloud storage service, SSDB – a hybrid SQL/NoSQL database service, SSBDS – a big data service, and SSVCS – a virtual cluster service. In this talk, we will focus on the design of SSBDS. The design of SSBDS is to integrate hardware and software to provide a platform, called HPBDA, for big data access, analysis, process, and presentation. The meanings of the term “high-performance” are two-fold. First, the platform equips high-performance hardware devices in storage (SSD), networking (InfiniBand), and computing (GPU). Second, the platform provides an enhanced and optimized Apache Hadoop software stack to achieve satisfactory performance for different big data applications. To reach this goal, we have developed (1) an enhanced HBase with multi-tenancy, fault-tolerance, transaction, and caching mechanisms; (2) an enhanced MapReduce framework with run-time optimization and caching mechanisms; and (3) an enhanced HDFS with heterogeneous storage, multi-tenancy, fault-tolerance, and global address mechanisms. Several data applications, such image processing and recognition, social network, and wafer fabrication, etc., are used to verify the design of SSBDS. |
