Research

SDN-enhanced MPI

Current interconnects of high-performance computing systems are static and do not reflect the actual communication pattern of applications. Our lab aims at automatically analyzing the communication characteristic of applications and dynamically reconfiguring the interconnect. Based on this idea, we specifically focus on how to accelerate various functions in Message Passing Interface (MPI), a de facto standard communication library for parallel and distributed programming.

Contention Avoidance of Staging Communication and Inter-process Communication

Computer cluster systems execute staging in which data used by a job is delivered to computation nodes used by the job beforehand. The conflicts that causes delays in communication time of jobs are happened because staging is executed even when computation nodes used by other jobs communicate with each other. In this research, we resolve this conflict by using Software Defined Networking in order to reduce communication time of each job and to improve computation time of that.

Visualization as a Service [web]

Tiled Display Wall (TDW) is one of the visualization system, which can provide a virtual large screen by using multiple display monitors. TDW is too expensive for most scientists to buy individually, and thus it is advisable to share TDW among scientists who need to use it. However, TDW has a problem that configuration for shared use is very difficult because scientists need a variety of visualization middleware on TDW. To solve this problem, we are developing a switching mechanism for visualization middleware.

OpenMN

Along with the requirements for network becoming increasingly diverse in recent years, it is necessary to make effective use of the network built with limited cost. However, making effective use of the network is difficult for software developers without knowledge and technique about networks. Open MN (Open Multi-Networking), our proposed novel programing model, allows software developers to use network resources effectively by using explicit and simple programming directives to describe network requirements.

Role-based network access control system

Frequent leaks of information assets at companies have been reported recently. In order to protect valuable information assets from security risks, security measures must be taken at each computers; however, a system to shut out unauthorized access in the network could add another level of security. Such system should be implemented in accordance with the organizational structure of a company (departments, security clearance levels, projects, etc.) and allow fine-grained and flexible access control. Based on this idea, our laboratory has been developing FlowSieve, a role-based network access control system built on top of OpenFlow. This research is a collaborative work with TIS Inc.

Sharing Economy of Things

With the arrival of Internet of Things (IoT) era, diversity of IoT devices have been connected to the Internet. However, IoT services have never been popular because of the lack of a platform for sharing IoT devices. We tackle this problem by proposing a platform for sharing IoT devices based on the idea of Sharing Economy.

Application of Deep Learning to Orthodontics

In recent years, image recognition and natural language processing by deep learning such as convolutional neural network and recurrent neural network attract a lot of attention. In addition, researches which combine image recognition and natural language processing are reported. In our laboratory, we study techniques to apply deep learning to orthodontic treatment in cooperation with Osaka University Dental Hospital. For example, we study predicting necessity of treatment from oral cavity images and generating treatment plan from face images.

Job Scheduling