GRIDtoday Logo Hewlett-Packard

DAILY NEWS AND INFORMATION FOR THE GLOBAL GRID COMMUNITY / FEBRUARY 03, 2003: VOL. 2 NO. 5

( Previous Article )   ( Table of Contents )   ( Next Article )

Special Features:

AN INTERVIEW WITH JAPAN'S RENOWNED SATOSHI SEKIGUCHI
By Alan Beck, Editor-in-Chief

At SC2002 we caught up with Mr. Satoshi Sekiguchi, Director of the Grid Technology and Research Center (GTRC) at the National Institute of Advanced Industrial Science and Technology in Japan.

Mr. Sekiguchi is well known in the Grid Computing industry as a visionary leader and will be program co-chairman for the Global Grid Forum's March GGF7 program in Tokyo. We wanted to get his views on GTRC's role in grid development and the importance of networking technologies on grid applications.

GRIDtoday, what are the goals of GTRC? What are some current projects under way?

Mr. Sekiguchi: Since 2002 the Grid Technology Research Center (GTRC) has been established as a part of National Institute of Advanced Industrial Science and Technology (AIST) funded by Japanese government. The primary objective of GTRC is to perform research on Grid technology in regards to areas of high- performance computing, networking, database, and their applications.

The GTRC also aims at becoming the focal point of research and development in the grid communities in Japan and Asia-Pacific region. We would accelerate the dramatic advancement and systematization of grid technology focusing on the development of the newest grid middleware, building of the test bed, and development of verification system by utilizing a high-performance computing system and large-scale databases. Currently, we have a couple of projects underway which include:

Ninf -- A grid middleware project which allows users to access remote computers using the proposed GGF standard "Grid RPC". Currently, we have provided an updated version of Ninf-G which is specifically designed for using standard grid security and information service.

ApGrid -- This started as an open community for collaboration among institutes, organizations, and individuals in Asia Pacific Region. Currently in this project we are developing an international Grid test bed and providing a venue for sharing information and exchanging knowledge as well as encouraging application communities to use Grid as a new computing infrastructure.

Gfarm -- Short for Grid Datafarm, Gfarm is designed for global petascale data- intensive computing. It provides a parallel cluster-of-cluster filesystem with online petascale storage, scalable I/O bandwidth, and scalable parallel processing, by exploiting local I/O in a grid of clusters with tens of thousands of nodes.

GRIDtoday: What are the main industrial drivers of grid technology in Japan?

Mr. Sekiguchi: It is a little bit difficult to identify who are real and active commercial players at this stage, every IT vendor is highly interested in grid technology. Definitely major Japanese IT vendors such as Fujitsu, NEC, and NTT group are deeply involved as well as major international companies like HP, IBM, Sun Microsystems, and Platform Computing.

GRIDtoday:How is GTRC/AIST working with other Grid Research centers globally to advance the state of grid technology?

Mr. Sekiguchi:At the global level, we fully respect the whole the Global Grid Forum (GGF) international activities which provide a body of standardizing grid technologies, a venue to exchange information, and more. In fact, GTRC is one of the gold sponsors of the GGF, and I have been serving as one of Steering Group members since GGF was founded.

Furthermore, GTRC has been leading the ApGrid initiative which coordinates grid activities among those countries in Asia-Pacific region to produce an international grid test bed. We support the NSF Middleware Initiative (NMI), Middleware and Grid Infrastructure Coordination Committee (MAGIC), and GRIDSTART and other coordinating bodies to realize international interoperability of Grids.

GRIDtoday: What brought you to SC2002?

Mr. Sekiguchi: The SCxx conference series are a showcase of the most advanced supercomputing and networking technologies both from both academic and business sectors. At the conference this year we had great opportunities to meet new people to explore possible future business plans together. Also, its networking experience with high bandwidth across the Pacific has brought us nicer inspirations.

GRIDtoday: Please tell us about your participation in the SC2002 Bandwidth Challenge. What application, CPU, network considerations did you have?

Mr. Sekiguchi: Please visit a couple of web site where you will see more info regarding the challenge.

Gfarm: http://datafarm.apgrid.org.

Grid Today: http://www.gridtoday.com/breaking/310.html.

Each 1U node of our 12-node grid cluster brought to the SC2002 research booth had a RAID that performs more than 120 MB/s of disk I/O bandwidth. When the disks perform full capability, we are easily limited by 1Gbps ceiling. In fact, we designed the cluster to fill up a 10Gbps connection when replicating file segments distributed among the cluster node.

Thus, 10Gbps is our minimum requirement. Due to budget limits, we didn't have 10G switches for this trial use, therefore we asked the SCinet committee to borrow one 10G box. The committee kindly asked Force10 Networks to contact us for this support. We knew Force 10 as one of the leading companies in the growing 10G network switch market, but unfortunately we had not had any experience with their product.

The Force10 switch showed the performance we wanted for our data grid solution, grid datafarm, and finally brought us success in our bandwidth challenge.

GRIDtoday: How does Ethernet and 10 Gigabit Ethernet in particular, fit into your current Grid computing architecture?

Mr. Sekiguchi: Ethernet provides high performance inner-cluster communication, and is interoperable with wide area networks. Since Ethernet is a commodity network, its cost is relatively low compared with other special networks, and will be even lower in the future. 10 Gigabit Ethernet is especially promising. It provides high bandwidth for replicating large-scale data, and makes it possible to share the large-scale data efficiently.

Also, we are looking at the future grid fabrication where clusters and grid are fully integrated by using commodity networking technology which is Ethernet.

GRIDtoday: How will Ethernet/10 Gigabit Ethernet help to speed the use of grid technology in industrial applications?

In industrial applications such as bio-technology and nano-technology applications, high-performance computing environments are required. Grid is a promising method to provide such an HPC environment.

Grid Clusters are key components to construct a Grid for HPC, and Ethernet (including GbE and 10GbE) is an important interconnection technology for inner and intra grid clusters communication. 10GbE provides a broad bandwidth connection between a cluster and a wide area of network, and will be used for the wide area network itself.

( Top of Page )

( Previous Article )   ( Table of Contents )   ( Next Article )