Computer Architecture for the Use and Application of Big Data

The Matsutani Lab in the Department of Information and Computer Science of Keio University's Faculty of Science and Technology is researching interconnection networks on a variety of scales ranging from large-scale data-center networks for Big Data to micro-networks such as Network-on-Chips.

"At this lab, we're currently conducting research mainly in four areas on computer architecture in order to fully utilize Big Data. The first area is research to accelerate databases so that Big Data can be utilized. More specifically, we're conducting research to accelerate simple but highly-scalable structured storages called NoSQLs using custom hardware."

At the Matsutani Lab, researchers are using FPGA boards that equip four 10 gigabit Ethernet interfaces in order to boost the database processing, and they are developing custom hardware for various structured storages. Human characteristics captured by a commercial facial-recognition system can be stored in the databases, which are accelerated by the proposed hardware.

"Our second area of research is social graphs, or in other words graphs that represent how individuals are connected. In this area, our research is directed at using GPUs to accelerate the analysis of social graphs such as Facebook. The third area has to do with the various requests that come to servers when dealing with Big Data and cloud computing. These requests that are received by servers have request-level parallelism, and in order to process numerous requests in parallel, many-core processors, which integrate many processing cores on a chip, must be used. At our lab as well, we're conducting research into many-core processors that integrate many processing cores, and more specifically, we're researching and developing wireless 3D many-core architecture that stacks many-core processor chips wirelessly."

At data centers that process Big Data, the cost to transfer data is becoming larger than ever before. Therefore, as the fourth area of research, the Matsutani Lab is conducting research into ultra-high speed networks that connect server racks by optical beams so that these types of data centers can handle unexpectedly large volumes of data transfers.

"Until now, these types of large data transfers were done with links of about 1 giga bps, and as a result, the data transfers took a tremendous amount of time. Therefore, at our lab, we're conducting research that will enable us to dynamically form optical beam links of 40 giga bps between racks by installing an element called a collimator lens on each server rack and then changing the direction of these collimator lenses. By doing this, immediately before a large volume data transfer is generated like the migration of virtual machines at a data center, a 40 giga bps wireless, or in other words, an optical wireless beam link can be formed, and then that optical beam can be used to transfer the large volume of data. That's the type of research we're conducting in this area."

The Matsutani Lab is conducting research on system architecture for Big Data from a computer architecture level, and is attempting to realize a computer infrastructure that will better facilitate the use and application of Big Data by achieving technology that conserves power, conserves space, and reduces costs. In addition, the lab intends to continue its research efforts into the future as well in order to support the major innovations that are expected in Big Data processing and analysis.