| 研究課題/領域番号 |
16F16349
|
|---|
| 研究機関 | 名古屋大学 |
|---|
研究代表者 |
加藤 ジェーン 名古屋大学, 情報科学研究科, 准教授 (70251882)
|
|---|
| 研究分担者 |
QI HENG 名古屋大学, 情報科学研究科, 外国人特別研究員
|
|---|
| 研究期間 (年度) |
2016-11-07 – 2018-03-31
|
| キーワード | Multimedia Big Data / Nearest Neighbor Search / SDN / Hadoop, Spark |
|---|
| 研究実績の概要 |
This project is about the collaborative management and control mechanisms of computing and networking resources in multimedia big data applications. In this year, the research achievements include three aspects. Firstly, we focus on the nearest neighbor search for large scale high-dimensional multimedia data, which is a fundamental problem in a variety of multimedia big data applications. We propose a novel grid based index structure and KNN graph based search algorithm. We also implement our algorithm on Apache Spark. Secondly, we focus on the SDN based big data platform, in which SDN can be used to manage the network resources efficiently, thereby improving the performance of big data processing. We propose an efficient forwarding scheme named Arbitrary Jump Source Routing (AJSR), which makes use of MPLS-based source routing. Thirdly, we build the prototype of multimedia big data platform. We have bought four Mac Mini PCs to build the environment of Hadoop for big data processing. In this year, we have published four papers, in which one is a SCI journal paper while other three are International conference papers.
|
| 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
This project is conducted in an orderly manner based on the research plan. In the proposed plan, we focus on the key problems of incorporating SDN into the big data platform. In this year, we address the routing issue in large scale interconnection network. We design and implement an efficient forwarding scheme named Arbitrary Jump Source Routing (AJSR). By AJSR we can achieve a trade-off between the control traffic overhead and the bandwidth overhead by dividing the complete routing path of a particular flow into arbitrary length sections and distributing these sections at different switches along the flow's routing path. This work belongs to the infrastructure part of the multimedia big data platform. Moreover, to finish a complete architecture of multimedia big data platform, we also focus on the key problems of multimedia big data applications. We choose large scale image search as an example. We propose new index structure and algorithm for ANN search of large scale image feature data. This work belongs to the application part of the multimedia big data platform. In this year, the research work includes the above two parts. In this year, the expected goal is achieved. The research achievements include four papers published in international journal and conferences.
|
| 今後の研究の推進方策 |
Based on the proposed plan, the future research work will include two parts: the algorithms of multimedia big data analysis and the SDN based platform for multimedia big data applications. In the first part, we will continue to conduct the research work about the ANN search of large scale image search. The final goal is to design an image search application as a typical application of multimedia big data processing. In the second part, we will focus on the SDN based collaborative management schemes of computing and networking resources in multimedia big data platform. The main work includes the application flow classification, the cost-performance tradeoff, and the SDN based network scheduling and management.
|
