APAN Grid Working Group Activities Report ======================= Youki Kadobayashi Nara Institute of Science and Technology January 31, 2001 This activities report is intended to compensate the WG chair's absense at APAN meeting in Hawaii. Question regarding this WG's activities can be answered on-site by Susumu Date or Kento Aida . In this report, we attempt to summarize past activities, existing projects, some wishes to APAN NOCs and short-term work items. 2. Projects 2.1. Telemicroscopy project NPACI's Telescience project (led by Prof. Mark Ellisman, UCSD/NCMIR) has been using the Telemicroscopy system developed by Osaka University and Hitachi, to conduct scientific observation and analysis of thick biological sections. The 3MeV electron microscope at Osaka University has been remotely controlled from the other end of the Pacific ocean, at the UC San Diego. Analysis of thick biological sections (e.g., microstructure of nerve cells) has been successfully conducted with latest technologies like IPv6 and Digital Video. It should be noted that this is not a one-time demonstration; the infrastructure has been used many times since 1999. The same research group at NPACI also demonstrated its flexible and more Internet-friendly telemicroscopy system that they have developed, at the INET'2000 conference in Yokohama, Japan. The group at NPACI joined forces with experts from NAIST to make it one of the first IPv6-enabled real application. Sun microsystems provided preliminary version of IPv6-enabled Java, specifically for this demonstration. Digital Video over IPv6 has also been used in this particular demo. 2.2. Grid-MEG project Research group at Osaka University (led by Dr. Yuko Mizuno-Matsumoto) has been successful in accelerating the analysis of brain function through MEG (magnetoencephalography). The group used Globus to reduce the time required to compute propagation delays of very weak magnetoelectric activities that are observed through MEG's 64 superconducting sensors. While their Wavelet-based analysis algorithm can preserve spatio-temporal characteristics, the computational cost was prohibitive in uniprocessor systems. Since they parallelized at multiple levels --- wavelet computation kernel has been parallelized, and cross-correlation analysis has been made data-parallel --- they expect they can achieve Teraflops levels of performance from massively parallel Grid infrastructure, once it becomes available. They have recently started collaboration with Institute of High Performance Computing at Singapore to further reduce the analysis time by having more computing resources within Asia-Pacific region. Their initial study at small-scale testbed indicated good scaling property. The group is also collaborating with Center for Signal Processing at Nanyang Technological University in Singapore. They are seeking wavelet bases and analysis algorithms that are more suitable to MEG signal's characteristics. 2.3. Ninf-NetSolve interoperability Ninf research group at Tokyo Institute of Technology (led by Prof. Satoshi Matsuoka) has been successfully collaborating with NetSolve researchers at University of Tennessee at Knoxville, using APAN/TransPAC as its underlying infrastructure. The two groups have come up with similar ideas of remote computing and initially developed different toolkits. Afterwards, they made two remote computing services interoperable. This was perhaps the first attempt to bring the idea of interoperability to Grid toolkits. Today, the notion of remote computing has been accepted in Grid Forum, the standardization forum of Grid technology, potentially making its way into one of their future standards. 2.4. PACX-MPI demonstration ETL research group (led by Dr. Satoshi Sekiguchi) has participated in the demonstrations of PACX-MPI at SC'2000 conference. HPC sites across Europe, US and Japan has been connected with PACX-MPI, which makes it possible to extend MPI computations to remote sites. PACX-MPI has been developed by HLRS in Germany (formerly Supercomputer Center at University of Stuttgart). 3. Requirements to HPRENs Since high performance research networks are engineered individually, any attempt to get reasonable bandwidth out of multi-national, multi-agency collaboration turns out to be major socio-technical challenge. While individual initiatives exist to ensure performance within single HPREN (e.g., I2 E2EPI), scientists want the same level of performance and the same set of advanced features across multiple HPRENs (e.g., stable and fast IPv6 connectivity across I2 - StarTAP - APAN - WIDE). While this problem is not unique to APAN, it should be made more aware among most of APAN's technology working groups. If HPRENs are going to form the core of the next generation Internet, they must interoperate at some of advanced (but well-established) features -- such as multicast, diffserv, measurement, and IPv6. The current array of HPRENs do not offer any of these features if we try to go from APAN to elsewhere. Most of applications reported above require QoS, since remote computations are sensitive to delay. Some of applications could use Diffserv together with IPv6, but we had to give up since the necessary infrastructure is not in place at some point in our path. If HPRENs have those advanced features in common and they are made interoperable, Grid toolkits will quickly make those features available to application programmers. 4. Work Items The idea of having Grid hands-on workshop in AP region has been discussed at the Beijing WG meeting. Many of WG attendees expressed intent to participate in the hands-on workshop. Research groups at Tsukuba are planning to host this workshop in mid-March. Participation to Global Grid Forum has been discussed many times. During INET'2000 period, GGF folks were seeking regional chapters, much like Europe's eGrid, but formation of regional chapter has not been mandatory. While the idea of "apGrid" has been tossed up many times, it still does not exist; perhaps because Grid communities in AP region are not yet well connected. Also, since Grid is a convergence of both high performance networking and high performance computing, Grid community does not exist in some countries. While AP region has very few things in common with Europe, it is always good to foster communities (and stimulating environment for students) among AP region, using Grid technology as one of the common vehicle. 5. Summary Grid has many aspects, not just technology development and deployment, but also scientific applications and technology sophistication for real applications. As such, our activities at APAN Grid WG have been diverse, broadly scoped and technologically shallow in some occasions. Therefore, we do not have any single tangible result that we can confidently say we have accomplished, from technological perspective. But our diverse activities have lead to strong requirements engineering and technology consolidation efforts, some of which will be reported later this year. Running advanced scientific application on top of advanced technology opens up tremendous opportunities for technology sophistication, well ahead of massive deployment in the commodity market. We realize the importance of Grid (as both technological driving force and reasonably good vehicle for science) and continue to work on it. This working group will continue to represent Grid activities as well as networking requirements to APAN/TransPAC networking professionals. 6. Upcoming events and deadlines 2001/2/16 Application deadline of SC Global constellation site 2001/3/3 HPDC-10 abstract submission deadline 2001/3/5-9 GGF1 and DataGrid workshop at Amsterdam 2001/3/15-16 Grid Hands-On Workshop at Tsukuba (tentative) 2001/4/27 SC2001 proposal deadline 2001/8/7-10 HPDC-10 2001/11/10-16 SC Global