“HKU Grid Point” Progress Report Nov. 28, 2004 The “HKU Grid Point” project started on Oct. 24, 2004. The System Research Group (SRG) of HKU and the research group in NCIC are in charge of this project. During the past two years, we made a steady progress and accomplished substantial achievement in the project. We reports the progress the SRG have made as follows: I. Building the Grid Platform The SRG built the “Gideon 300 Cluster” in the end of October in 2002. The Gideon 300 Cluster consists of 300 Intel-based Legend PCs interconnected by a single, high-port density Fast-Ethernet switch. This guarantees a one-hop latency between each node with full connectivity. Each processing node is a standard PC which comprises an Intel Pentium 4 processor running at 2 GHz, a 512 Mbytes (PC2100) DDR SDRAM, and a 40GB IDE hard disk. This gives the whole cluster a peak floating point performance of 1.2 Tflops (trillion floating point operations per second). The cluster has been benchmarked according to an international standard - the High-Performance Linpack benchmark. The Gideon 300 Cluster is ranked #175 on the TOP500 Supercomputers list for November 2002. We installed GOS software and the portal in 32 machines of the cluster and register the resource to the server in Computer Network Information Center. Now the user can get the information of the resource and grid services provided by HKU Grid Point. II. Grid Services Through the portal of China National Grid, users can access various grid services provided by HKU grid point. The following services are produced from some R&D projects carried out by our group: JESSICA2 service: It provides the service of parallel Java virtual machine. Users can submit their own Java multi-thread sequential program. The multiple threads can be distributed to multiple machines to be executed in parallel. The efficiency is much higher than that in one single machine. The service returns the output back to the user through the portal. G-JavaMPI service: It provides the JavaMPI programming and execution environment. Users can submit their own JavaMPI programs. The service returns the output back to the user through the portal. The other services are produced from software released by other research groups: WireGL service: WireGL is a cluster rendering systems developed by Stanford University. The software allows graphics applications to render to a cluster of workstations outputting to a tiled display. The service provides the service of 3-D graph or animation rendering. MatMPI service: It uses the software “MatlabMPI” developed by MIT which allows any Matlab program to be run on a parallel computer and provides the service of parallel mathematic computation. Drug Discovery service: Our site is one of grid points of Shanghai Drug Discovery Grid which is developed by some institutions in Shanghai. Our Gideon cluster provides the computational resource to execute the task of drug discovery, and returns the results to the server in Shanghai. III. Grid R&D Projects G-JavaMPI: This project does research on the grid middleware providing simple but efficient grid programming and execution environment. We inroduce a grid middleware called G-JavaMPI, which combines a high-level message passing interface with the Java language to support portable messaging-passing programming in a grid. Different from traditional MPI implementations, it supports transparent migration of MPI processes during execution. This feature facilitates more flexible task scheduling and more effective resource sharing. The migration mechanism is implemented by exploiting the JVM Debugger Interface (JVMDI) functions, and minor bytecode modification. This method is portable and does not require modification of the JVM. To guarantee continuous MPI communication during process migration, a message redirection mechanism is employed. Based on G-JavaMPI, we develop a BLAST application called “JmpiBLAST” to evaluate the performance of MPI communication and process migration. We also investigate the dynamic task scheduling algorithm for this application and other similar applications. During runtime, when there is new resource joining in or quitting, the algorithm calculates the new mapping of the processes in response to the changing. The results attest to the effectiveness of the algorithm. The detailed experimental results are reported in the published papers. Currently G-JavaMPI and JmpiBLAST are two grid services provided in the portal that can be accessed by the users. G-PASS: This research focuses on providing flexible security support for mobile agents and processes migrating frequently between organizations in grid. The main idea includes an instance-oriented authorization mechanism and a role-based inter-organization mapping. G-PASS simulates the custom system for international immigration management in real world. An unforgeable document named G-passport is proposed as the kernel of the security supporting system. Several relative protocols are also defined. G-PASS can be an infrastructure used in deploying the grid-based mobile applications. It is completely compatible with the credentials and certificates in GSI, and more powerful supporting functionality is provided. G-PASS has been configured in G-JavaMPI to support for the authorization procedure during process migration. Experiments show that the G-PASS works very well in the collaboration and delivers acceptable overhead. InstantGrid: This research proposes a framework for on-demand construction of grid points. In contrast to traditional approaches, InstandGrid is designed to substantially simplify software management in grid systems, and is able to instantly turn any computer into a grid-ready platform with the desired execution environment. We experiment on the Gideon cluster, and the results demonstrate that a 256-node grid point with commodity grid middleware can be constructed in five minutes from scratch. Detailed information can be found in the paper published in the conference GCC04. Publications: Lin Chen, Tianchi Ma, Cho-Li Wang, Francis C.M. Lau, and Shanping Li, ``GJavaMPI: A Grid Middleware for Transparent MPI Task Migration,'' to appear in the book "High Performance Computing: Paradigm and Infrastructure", John Wiley and Sons, Spring 2005. Roy S.C. Ho, K.K. Yin, David C.M. Lee, Daniel H.F. Hung, C.L. Wang, and Francis C.M. Lau, ``InstantGrid: A Framework for On-Demand Grid Point Construction,'' The International Workshop on Grid and Cooperative Computing (GCC 2004), pp. 911-914, Oct 21-24, 2004, Wuhan, China. Tianchi Ma, Lin Chen, Cho-Li Wang, and Francis C.M. Lau, ``G-PASS: Security Infrastructure for Grid Travelers, The International Workshop on Grid and Cooperative Computing (GCC 2004), pp. 301-308, Oct 21-24, 2004, Wuhan, China. Lin Chen, Cho-Li Wang, Francis C.M. Lau, and Ricky K. K. Ma, ``A Grid Middleware for Distributed Java Computing with MPI Binding and Process Migration Supports,'' International Workshop on Grid and Cooperative Computing (GCC-2002), December 26-28, 2002, Hainan, China, , pp. 640-652. An extended version of this paper has appeared in Journal of Computer Science and Technology, Vol. 18, No. 4, July 2003, pp. 505-514 IV. Aspects to Be Improved The grid services and the portal are not steady enough. Some research projects are in experimental level. V. Future Work Improve the portal and grid services Research on efficient dynamic scheduling algorithms based on process migration for various grid applications. Promote the resource sharing and information exchange between CNGrid and HKGrid