| 研究課題/領域番号 |
15K00428
|
|---|
| 研究機関 | 会津大学 |
|---|
研究代表者 |
Paik Incheon 会津大学, コンピュータ理工学部, 准教授 (70336478)
|
|---|
| 研究期間 (年度) |
2015-04-01 – 2018-03-31
|
| キーワード | Service Discovery / Social Service Network / Service Graph / Distributed Processing / Big Data / Hadoop / Graph Partition / Map-Reduce |
|---|
| 研究実績の概要 |
Service discovery and composition are challenging issues in service computing to provide value-added service. Existing approaches using keyword or ontology matching have limitations for realistic service discovery and composition, considering nonfunctionality or sociality. One main reason for these limitations is that the approaches are based on isolated services. The efficient discovery and composition of services is hindered by this isolation. Therefore, past research suggested linked social service networks considering relationships between functional and nonfunctional properties, and social interaction based on complex network theory, where related services can be located through sociability. However, it would become difficult to create a linked social service network because of the large latency in calculating link quality and increasing new services, such as portable devices and sensors, with the progress in big data technology. In this research, creation of a linked social service network to improve network construction performance by considering a distributed process using big data infrastructure has been proposed. First, an algorithm that creates a network graph using a MapReduce parallel programming model has been devised, and the experiment for performance of distributed computation and characteristics of generated service network have been conducted.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Global Social Service Network (GSSN) service registry in the form of social networks in public areas is very useful for locating services that a user wants using several attributes. In this research, we address the large cost for creating GSSN by parallel processing using Map-Reduce(MR) on Hadoop big data infrastructure. Issues and steps that have been accomplished are as follows.
First, confirmation of the existing GSSN and data set and decision of calculation of link properties on different environment was conducted. Second, development of method for splitting GSSN for distributed computation was done. Third, development of all procedure to calculate quality of link using Map-Reduce parallel computation and implemented it on Hadoop distributed clusters have been conducted. Finally, evaluation of performance to create MR-GSSN and characteristics of the generated graph were carried out.
The proposed algorithm creates MR-GSSN with the same network characteristics with the existing GSSN created by single machine. Computation time for MR-GSSN has been improved by 9.8–20 times (1ms – 3ms in delay) when we used 18 Hadoop data nodes, and increasing rate of computation performance for number of service improved by 16 times. The ratio of average increase in rates of computation slope by node delay and number of services between single computation and distributed computation is about 35 times, which shows that our proposed method works with high scalability and efficiently for GSSN construction in a distributed environment.
|
| 今後の研究の推進方策 |
Title for FY 2016: Improving Link Quality Calculation and Exploitation of GSSN for Better Discovery for Composition
In the second year, the basic formulation with minimal parameter set for calculating link quality and the previous GSSN will be improved. And methodologies of exploiting the GSSN to find related service will be developed.
1)Improving the link quality formulation: Calculation of QDSR(R, T) is affected by service similarity by data correlation, term similarity, and ontology matching. More complete service similarity will be devised such as ontology, search engine based method, and machine learning method to improve link quality. Also, for better computation performance, Apache Spark on HDFS will be tried and compared with the existing Hadoop Map-Reduce.
2)Methodology of Exploiting GSSN for Services Discovery for Composition: To develop algorithms to map the GSSN into a service cluster network to reduce the search space, and a quality-driven discovery approach is proposed to enable exploitation of the service cluster network, considering composition workflow.
3)GSSN service construction and evaluation: More complete and improved GSSN system will be constructed, and experiment of publication of services on the GSSN will be conducted. Experiment and evaluation with various parameters in the link quality formula and the service complex network for the objective function of better composition quality. Also, performance comparison with Big Data approach will be conducted.
|
| 次年度使用額が生じた理由 |
The 1 yen has been remained as there was no such a goods with small price.
|
| 次年度使用額の使用計画 |
I will use the 1 yen to buy a consumable goods to add it to FY 2016 budget.
|
