Presentations—October 2004 Gelato Federation Meeting
Nearly 60 scientists and engineers from 30 Gelato Member and Sponsor institutions attended the October 2004 Gelato
Federation Meeting, which focused on current high-performance computing issues and collaborative solutions specific to the Itanium platform. Gelato would like to thank Tsinghua University, our meeting host, for a great event with well-planned social events and carefully-managed logistics. In a three-day period, attendees were treated to presentations by some of the top research and industry users of Linux on Itanium. Technically-rich presentations included: Chung Shin Yee's Cache-Oblivious Algorithms, Peter Chubb's Linux Scalability, Research on IA-64 at Tsinghua University from Weimin Zheng's group, Andreas Hirstius' High-Speed Data Transfers, and Bob Kidd's Overview of OpenIMPACT. Be sure to check out these presentations and more.
Besides the presentations listed below, view some of the photographs from the meeting.
| | | | Day One - Monday, October 11 | Research on IA-64 at Tsinghua University (PDF)
| Weimin Zheng and Team (Tsinghua University)Weimin Zheng is President and Professor of the Institute of High Performance Computing in the Department of Computer Science and Technology at Tsinghua University in Beijing, China. His major research interests include parallel/distributed and cluster computing, compiler techniques, and run-time system design for parallel processing systems. Zheng and his group—faculty, research staff members, and graduate students—are currently working on a number of R&D projects supported by the Natural Science Foundation of China and the National High-Tech Program. Zheng has also been engaged in many important projects, including the National High Technology Research and Development Program of China (863 Program) and is the Managing Director of the Chinese Computer Society. | | | THDRMS—A Digital Resource Management System Based on Open-Source Systems (PDF)
Digital libraries manage very large volumes of text documents and digitized materials such as pictures, audio, video, text, and e-books. This management requires a platform to provide functions to store and index the document materials. In addition, it must support information services to end-users. This report reviews the requirement and then introduces a platform called THDRMS. The platform has multi-tiered architecture and is built on open-source systems such as Linux, Fedora, and MySQL. Using THDRMS, two digital libraries have been developed. | Lizhu Zhou (Tsinghua University)Lizhu Zhou is a Professor of the Department of Computer Science and Technology at Tsinghua University in Beijing, China, and the Chair of the Department Academic Committee. He received his MS from the University of Toronto in 1983. His major research interests include database systems, digital resource management, Web data processing, and information system design. Since the 1980s, Professor Zhou has led his group to complete over 20 R&D projects funded by the National Natural Science Foundation of China and China~Rs high-tech programs and industries. He has published over 70 papers in journals and conferences and has served as a PC member, Program Co-Chair, or General Chair for many international conferences. | | | Smart Data Center (PDF)
The Smart Data Center is "smart" by design and operation and adapts to the business needs of the customer. But the state-of-the-art data center design fails to achieve the compaction and data center consolidation sought by customers. Limitations in removing high heat loads of 2000-3000 W/m2 often inhibit the creation of high-performance, standards-based supercomputing clusters. Furthermore, inflexibility in components, proprietary interfaces, and protocol result in inefficient and labor-intensive operation. By contrast, Smart Data Center design tools enable efficient provisioning based on requirements (capacity, availability, performance). The Smart Data Center can dynamically adapt to changes in business needs and exception events in the infrastructure so that requirements can be met as cost effectively as possible. HP has built a prototype Smart Data Center in Palo Alto, California, to demonstrate customer benefits, such as more efficient use of energy, space, and resources. This talk will present the thermo-mechanical instantiation of the Smart Data Center and results gathered to date. | Brian Lynn (Hewlett Packard)Brian Lynn is the Manager of the Linux Technologies project at HP Labs. He is the HP Labs technical liaison to the Gelato Federation. | | | Openlab—High-Speed Data Transfers (LAN/WAN) (PDF)
The main focus of this presentation will be high-speed wide-area networking with a short introduction into the issues and challenges of wide-area (transoceanic) networking. The latest measurements of memory-to-memory and disk-to-memory transfers will be presented, along with a general overview of the openlab activities. | Andreas Hirstius (CERN)Andreas Hirstius is a CERN openlab Fellow. He has been working on CERN experiments since 1998, first on the L3 experiment and later on the NA48 experiment. At the latter, he was responsible for the third-level trigger and central data recording. He analyzed data from this experiment for his PhD thesis. His current fields of interest are hardware related and include testing of 10Gb networking hardware (Ethernet and InfiniBand) and advanced storage systems. He is also involved in a 10Gb WAN PHY long-haul test, the latest DataTAG measurements, and high speed long-haul data transfer. | | | Challenges for Linux with Large and Shared Filesystems (PDF)
| Doug Johnson (OSC)Doug Johnson is the Technical Lead for the Cluster Ohio Project and the production Linux clusters at OSC. He has worked on many projects to address usability and manageability of clusters of commodity systems. His current areas of interest include: grid meta-scheduling, storage for clusters, and high-availability services for clusters. | | | Real-Time Semantic Composition of Grid/Web Services (PDF)
The concept of services composition comes from the area of Web services. It can well integrate existing Web services and form a more complex and powerful "big" service. There has been much research work on Web services composition already, but most is theoretical. Few application systems have been developed yet, and the issues of automation and semantic conflict are not well resolved. Therefore, we developed a Grid/Web services composition tool based on Globus Toolkit and BPEL4WS. This tool can easily describe workflow composition by BPEL4WS, and it can also use ontology to solve the semantic problem during the composition process. Thus, the Grid/Web services composition can be done fully automatically. As long as users give the input and the desired output, the system will automatically compose a new service workflow. Then, the execution engine will execute the process and return the result. Furthermore, when creating a composition plan, the system can choose certain services according to the user's SLA so that the composition can fully satisfy the user's QoS requirement. | Ning Gu (Fundan University)Professor Ning Gu is the Head of the Department of Computing and Information Technology at Fudan University in China and is also the Director of the Collaborative Information and Systems Lab. He got his PhD at the Institute of Computing Technology at the Chinese Academy of Sciences in 1995 and did his post-doctoral research at Fudan University from 1995 to 1996. His research interests include Grid computing, CSCW, and data and knowledge management. In 2002, he received the esteemed Shanghai Scientific Progress Award. He and his group are now working on several projects funded by the National Natural Science Foundation of China and the Shanghai Science and Technology Committee, including virtual organization technologies in grid, Web services composition, and grid virtual data storage. | | | Design and Grid Application Development of Mixed Linux Clusters (PDF)
This presentation focuses on the design and implementation of middleware systems, as well as on the applications developed on SCUT's mixed-grid computing platform, which comprises several heterogeneous cluster systems, one of which is a Linux Itanium2 cluster. A scheduler for the mixed Linux cluster will be proposed. Some applications, such as parallel search engines and mpiblast, have been designed and tested in the heterogeneous environments. Future research on developing a parallel distributed application will also be discussed. | Ling Zhang (SCUT)Ling Zhang is a Professor in the School of Electronic and Information Engineering at the South China University of Technology (SCUT), where he specializes in communication networking and high-performance computing. Prof. Zhang received his PhD in Communication and Information Systems from the South China University of Technology in 1989. He was a post-doctoral researcher at Oxford Brooks University and a visiting scientist at the National Rutherford Research Lab, United Kingdom. Currently, Prof. Zhang is the Dean of the School of Computer Science and Engineering at SCUT, a Deputy Director of the expert committee at the China Research and Education Network (CERNET), a key member of the expert committee of the National High Technology Research and Development Program of China (863 Program), and one of five ChinaGrid founding members. | | | Grid Computing and the ChinaGrid Project (PDF)
Grid computing presents a new trend in distributed computing and Internet applications, which can construct a virtual, single image of heterogeneous resources, provide a uniform application interface, and integrate widespread computational resources into a super, ubiquitous, and transparent aggregation. The ChinaGrid Project, founded by the Ministry of Education of China, is an attempt to achieve these goals by exploring the various resources of the existing and well-developed Internet infrastructure, CERNET (China Education and Research Network). In this talk, I will introduce a general picture of the ChinaGrid Project, along with its vision and mission. The design of the ChinaGrid Support Platform, CGSP, will also be briefly discussed. To illustrate the reality of the ChinaGrid Project, five different grid computing applications and their application supporting platforms will be discussed in detail. | Hai Jin (Huazhong University)Hai Jin is a Professor of Computer Science and Engineering at the Huazhong University of Science and Technology (HUST) in China. He received his PhD in Computer Engineering from HUST in 1994. In 1996, he was awarded the German Academic Exchange Service (DAAD) Fellowship at the Technical University of Chemnitz in Germany. He worked for the University of Hong Kong between 1998 and 2000 and participated in the HKU Cluster Project. He was a visiting scholar at the University of Southern California between 1999 and 2000. He is the chief scientist of ChinaGrid, the largest grid computing project in China. Dr. Jin is a member of IEEE and ACM. He is the Associate Editor of "The International Journal of Grid and Web Services," "The International Journal of Computer and Applications," and "The Journal of Computer Science and Technology." He is the executive member and region coordinator of the IEEE Task Force on Cluster Computing (TFCC). He is a technical committee member of the International Association of Science and Technology for Development (IASTED) for parallel and distributed computing and systems. He is the steering committee member of the IEEE/ACM International Symposium on Cluster Computing and the Grid (CCGrid), and served as a program vice-chair of the CCGrid'01, PDCAT'03, and NPC'04; and a program chair of GCC'04. He served as conference chairs for the International Workshop on Internet Computing and E-Commerce from 2001 to 2003. He also served on program committees for more than 50 international conferences and workshops. He has co-authored four books and published over 120 research papers. His research interests include cluster computing and grid computing, network storage, network security, and fault tolerance. | | | Projects MASSIVE and DART (PDF)
| Yao Zheng (Zhejiang University)Yao Zheng is a Cheung Kong chair professor with the College of Computer Science, Zhejiang University, appointed by the Ministry of Education of China since 2001. He is also the director of the Center for Engineering and Scientific Computation at Zhejiang University. He had been a Senior Research Scientist for Taitech, Inc., NASA Glenn Research Center, Cleveland, Ohio, from 1998 to 2002. Prior to moving to NASA, he had been a Senior Software Scientist with Analysis and Design Application Company (adapco), New York for two years; and a Senior Research Assistant (Research Associate) at the Department of Civil Engineering (Institute for Numerical Methods in Engineering), University of Wales Swansea, UK for six years. He earned his PhD in Computational Modeling from University of Wales. Dr. Zheng has been awarded the 2002 National Science Fund for Distinguished Young Scholars, by the National Nature Science Foundation of China. He is currently leading a group working on projects such as EEMAS (Enabling Environment for Multidisciplinary Application Simulations), and MASSIVE (Multidisciplinary ApplicationS-oriented SImulation and Visualization Environment) for computational grids. His primary interests include high performance computing, and enabling technology for multidisciplinary application simulations. | | | Research at BMD (PDF)
| Kyoung Tai No (Research Institute of Bioinformatics & Molecular Design)Kyoung Tai No, PhD, is the Director of the Research Institute of Bioinformatics & Molecular Design (BMD), working in the Department of Bioengineering at Yonsei University. Dr. No's current research areas focus on: structure- and ligan-based drug design, drug design using virtual screening [anti-viral, anti-fungal, anti-cancer, QOL (quality of life)], ADME/Tox (adsorption, distribution, metabolism, excretion/toxicity) prediction, and the construction of several databases, including: organic chemical databases for drug design and nano material design, a protein structure database, a pharmacophore database, and a physiological database for ADME/Tox. Dr. No is also a Professor at Yonsei University in Korea, and is a member of: the American Chemistry Society, the American Protein Society, and the PHYSIOME Society of Korea, and is a board member of the Korean Society of Bioinformatics. | | | Peking University's HPC/Linux/Itanium Related Activities (PDF)
| Xiaoming Li (PKU)
| | | | Day Two - Tuesday, October 12 | IA-64 Performance Tools (PDF)
Itanium processors have very sophisticated performance monitoring tools integrated into the CPU. McKinley and Madison Itanium CPUs can filter, trigger on, and count over 300 different types of events. The restriction on which combinations of triggers are allowed is daunting and varies across CPU implementations. Fortunately, the tools hide this complicated mess. While the tools prevent us from shooting ourselves in the foot, it is not obvious how to use those tools for measuring kernel device driver behaviors.
IO driver writers can use pfmon to measure two key areas generally not obvious from the code: MMIO read and write frequency and precise addresses of instructions regularly causing L3 data cache misses. Measuring MMIO reads has some nuances related to instruction execution which are relevant to understanding IA-64 and likely IA-32 platforms. Similarly, the ability to pinpoint exactly which data are being accessed by drivers enables driver writers to either modify the algorithms or add pre-fetching directives where feasible. I include some examples on how I used pfmon to measure NIC drivers and give some guidelines on use.
q-syscollect is a "gprof without the pain" kind of tool. While q-syscollect uses the same kernel perfmon subsystem as pfmon, the former works at a higher level. With some knowledge about how the kernel operates, q-syscollect can collect call-graphs, function call counts, and percentage of time spent in particular routines. In other words, pfmon can tell us how much time the CPU spends stalled on d-cache misses, and q-syscollect can give us the call-graph for the worst offenders.
There is also a white paper titled IA64-Linux perf tools for IO dorks that corresponds to this presentation. | Grant Grundler (Hewlett Packard)Grant Grundler was born in Toronto, Canada, and graduated from California State University, Hayward with a BS in Computer Science. He lived and worked in Germany as a PC technician/support, ski tour "host," windsurf instructor, and firmware designer/developer for a custom TokenBus networking card. Back in "the States," he worked for three years at Olivetti on SVR4 ports to i860, MIPS R4000 (M700-10), and the first Alpha workstation. Since 1993, Grant has worked for HP on HPUX SCSI drivers, re-architected HPUX PCI subsystem support, and the parisc- and ia64-linux ports. He wrote nearly all of the architecture-specific PCI, IO IRQ, and IOMMU support for the parisc-linux port. A fair portion of the PA-RISC PCI code was leveraged for HP ZX1 (IA-64) platforms. Like this presentation on IA-64 performance measuring tools, previous OLS presentations focused on issues related to developing device drivers: "Logic and Protocol Analyzers" (2001), "Porting Drivers to ZX1" (2002), and "DMA Hints on ZX1" (2003). | | | Overview of OpenIMPACT (PDF)
The OpenIMPACT team has made impressive progress in the past five months. With the first public release in September 2004, Gelato Members got their first look at the compiler. This talk will begin with an update on the past few months' development. The bulk of the talk will be an overview of the compiler from a user's point of view. While the compiler has been engineered to support a gcc-like use model, the limitations of this model prevent certain optimizations. This talk will demonstrate how to use OpenIMPACT's more advanced optimizations. | Bob Kidd (UIUC)Robert Kidd is a Compiler Engineer in the OpenIMPACT team at the University of Illinois at Urbana-Champaign (UIUC), where he is responsible for developing a new front-end library to support future compiler research. He has also worked on back-end enhancements, as well as scripting drivers for GCC compatibility and compiler usability. Robert has a BS in Computer Science with Highest Honors from UIUC.
Shailesh Patel (UIUC)Shailesh Patel was born in India, and grew up in Dubai, UAE. He graduated from the National Institute of Technology (NIT), India with a BS in Engineering and then completed his MS in Computer Engineering from California State University, Long Beach. He has worked as a J2EE developer, creating software for the subtitling and marketing industry. At UIUC, he has worked with the SandBox group and is now working with the IMPACT team. | | | Scalable Cluster Middleware: INRIA's Perspective (PDF)
| Brigitte Plateau (INRIA)Brigitte Plateau is the Project Manager for the APACHE parallel processing research project sponsored in part by the National Institute for Research in Computer Science and Control (INRIA) in France. Plateau was born in France in 1954. She received an MS in Applied Mathematics from the University of Paris 6, and a Thèse de Troisième Cycle and a Thèse d'Etat in Computer Science from the University of Paris 11. She was Chargé de Recherche at CNRS in France from 1981 to 1985, and an Assistant Professor at the Computer Science Department of the University of Maryland from 1985 to 1987. Plateau is currently a Professor at the Engineering School (ENSIMAG) in Grenoble, France, and is heading a research group whose main interest is parallel programming. Her research interests include modeling and performance evaluation of parallel and distributed computer systems, numerical solution, and simulation of large Markov models. | | | Work on Linux Scalability (PDF)
From the Linux Watch to 1024-plus processor NUMA machines, Linux is possibly the most scalable operating system *ever.* In this brief talk, I will give an outline of what scalability as applied to the operating system *is,* give a lightning-fast overview of current work in the field, and the most prominent problems. I will also try to give a feel for the kinds of things people programming in the kernel have to be aware of to allow good scalability. | Peter Chubb (UNSW)Peter Chubb is a Senior Research Engineer at National ICT Australia and a Research Officer at UNSW. He completed his PhD under Associate Professor John Lions in 1989. Peter worked at Softway Pty Ltd as a consultant and software engineer doing UNIX kernel, security, and embedded work. He joined Gelato@UNSW at its inception in 2002. Peter started using UNIX in 1979 and has never used Microsoft operating systems for more than a few moments. His home life includes wife Lucy, who also works at Gelato@UNSW, and two small daughters. Peter's hobbies include music (he runs a recorder consort), aquaria (3 tanks at present, no room for more), and fine wines. | | | Status Report of A Performance Test Run on a Itanium NUMA System (PDF)
| Ping-Hui Kao (Hewlett Packard)After working at Motorola on UNIX applications for two years, Ping-Hui Kao joined HP in 1984. Over the years, Kao has worked on HPUX utilities (vi, sh, ksh, and many others), HPUX kernel (file systems), HP-DUX distributed file systems, HPUX clusters, Windows NT on PA-RISC, Windows clusters, HPUX kernel research, HPUX network drivers, Linux Secure kernel, Linux clusters, Linux scalability, and system architecture. | | | | Day Three - Wednesday, October 13 | The Southern Partnership for Advanced Computational Infrastructure (PDF)
| Italo Epicoco (SPACI)
| | | Gelato Activities at the University of Copenhagen (PDF)
| Eric Jul (DIKU)
| | | Cache-Oblivious Algorithms (PDF)
| Chung Shing Yee (IHPC)Chung Shin Yee is a research officer at IHPC. His research interests include parallel computing, cache-efficient algorithms, and data structures. He started to work in the area of parallel computing during his internship with SGI in 2001 when developing parallel DNA sequence matching using OpenMP on an SGI Origin 2000. He received a BS in Computer Engineering from Nanyang Technological University in Singapore in 2003. During his final year of study, he did research on parallel DNA multiple sequence alignment using MPI. After his graduation, he joined the School of Computer Engineering at Nanyang Technological University in May 2003 as a project officer to work on Web information extraction. In October 2003, he started to pursue postgraduate study on cache-oblivious algorithms and data structures while working as a research officer at IHPC. |
|
