University of Houston 
 |
Housed at the University of Houston, the Texas Learning and Computation Center (TLC2)
serves as a campus-wide center to foster and support multidisciplinary research, education, and training in the computational sciences, computer science, and information technology.
Computational research is focused primarily in Physics, Chemistry, Biochemistry, Geosciences, Mathematics and Engineering.
Most of their codes are either community codes or developed within research groups at the University. TLC2 operates a visualization laboratory, high-performance computing systems, and
systems for computer science research, as well as a distributed storage facility with servers in several campus locations integrated with compute servers over a dedicated Gigabit Ethernet. TLC2
research groups include faculty and students in several Colleges, including Natural Sciences and Mathematics, Technology, Engineering, and Social Sciences. Research groups affiliated with TLC2 are
actively engaged in several national and international science, middleware, and tools projects, including Grid projects. Directly or through disciplinary member centers, TLC2 industrial affiliates
include: Amerada Hess, BP, Chevron-Texaco, CGG, Conoco-Phillips, ENI-Agip, Exxon-Mobil, GX Technology, Petrobras, Shell, Statoil, Total, Unocal, WesternGeo, CoreLabs, PGS, HP, and Intel.
|
People | Lennart Johnsson - Gelato Leader | 
|
S. Lennart Johnsson wears many hats, both inside and outside the Gelato Federation. Within the Federation, he is a Strategy Council Representative for two institutions: the Royal Institute of Technology (KTH) in Stockholm, Sweden and the University of Houston in Texas, USA.
At the Royal Institute of Technology (KTH) in Stockholm, Sweden, Dr. Johnsson is a Professor of Numerical Analysis and Computer Science and the Director of the Center for Parallel Computers (PDC) at KTH. At the University of Houston, he is a Hugh Roy and Lillie Cranz Cullen Distinguished Professor of Computer Science, Mathematics and Electrical and Computer Engineering, and is the Director of the Texas Learning and Computation Center (TLC2). Dr. Johnsson is also an Adjunct Professor of Computer Science, Rice University.
Dr. Johnsson's research interests include computational and data Grids; high-performance scientific computation; parallel algorithms; adaptive, Grid-aware, high-performance software and tools for the creation thereof; and middleware for Grids and parallel computers, especially communications-related middleware, performance modeling, and problem-solving environments. He actively contributes to establishing the European Grid Forum that now has merged with the U.S. Grid Forum to form the Global Grid Forum. Recently, he lead the formation of the Nordic Grid Consortium and the European Grid Support Center.
During his four decades in the technology sector, Dr. Johnsson has taught and researched at UCLA, the California Institute of Technology, Argonne National Laboratories, Yale University, Harvard University, Stanford University, Rice University, Baylor College of Medicine, the University of Houston, and several universities in Sweden and Norway. He is a mentor to Ph.D. and Masters Students, an organizer and panelist at numerous technology symposiums, a member of the board of a number of technology institutes in the USA and abroad, and an editor of seven computing and scientific journals. Dr. Johnsson can be reached at johnsson@cs.uh.edu.
|
| Dragan Mirkovic | 
| A Research Scientist at TLC2, Dragan Mirkovic is a visiting assistant professor at the Department of Computer Science at University of Houston. He received his Ph.D. in Applied Mathematics from the State University of New York in Stony Brook. His research interests include development of software and tools for high-performance scientific computing. |
| Bo Liu | 
| Bo Liu is a Ph.D. student in the Department of Computer Science at the University of Houston working in the Advanced Computing Research Lab (ACRL). She is currently working for Dr. Lennart Johnsson on resource management in Grid, including application performance modeling, sandbox and QoS in Grid. |
| Mitul Patel | 
| Mitul Patel is a Ph.D. student in the Department of Computer Science at the University of Houston. His research work involves performance analysis and modeling of scientific applications. In addition to the understanding of specific applications, it requires in depth study of operating systems, compilers and various computer architectures. He currently has a Masters degree in Electrical Engineering from the University of Houston. |
Project Overview
Grid is an emerging infrastructure to allow the flexible sharing and collaboration of all components of our information technology infrastructure. The core functionality to manage all these resources is to establish a mutual agreement between a resource provider and a resource consumer. The goal of the application performance modeling is to guide the resource consumer to decide how many resources his tasks will need in a certain Grid and help the scheduler to make the decision.
The goal of the application performance modeling is to guide the resource consumer to decide how many resources his tasks will need in a certain Grid and help the Grid scheduler to make decision.
Technical Approach
The performance model of a Grid application is a function of both its single node performance and how it exercises the network infrastructure. To estimate the execution cost of an application on arbitrary Grid configurations, we analyze an applications behavior by modeling its characteristics in isolation of any architectural details and then estimate the applications execution cost on a target platform described by its available hardware resources. Specifically, we use Memory Reuse Distance (MRD) to characterize the application memory access pattern and model the collected information about MRD into memory miss models. We use hardware counters and executable instrumentation to get the application computation load and model the collected data into computation models.
Results
We do single node performance modeling semi-automatically. We can characterize the application independent from the architecture and integrate the application models with the architecture models into the single node performance model for a certain target platform.
Looking Ahead
From the semi-automatic to automatic tool is our first priority in the future. The second is to improve the method for architecture independent determination of floating point operation count. Also, to incorporate the application control variables and compiler effects will be addressed in the future.
Equipment Used
TLC2's current Itanium 2 cluster of 61 nodes comprised of 60 HP zx6000 workstations and 1 HP rz5670. The interconnect is Dolphin SCI based and all the systems are currently running RedHat Advanced Workstation 2.1.
| 
TLC2's new 152 node Itanium 2 cluster currently being built in conjunction with HP. It is comprised of 152 HP rx2600 machines running RedHat Advanced Server 2.1 with a Myrinet interconnect.
| 
2's Myrinet M3-E128 switch and incoming fiber patches for the 152 node cluster.
|
Resources
Related Papers
 High Performance Numerical Components (pdf)
New Results with the UHFFT Library (pdf)
Automatic Performance Tuning for Fast Fourier Transforms, Dragan Mirkovic, Lennart Johnsson,
Department of Computer Science, University of Houston, Houston, TX 77204, USA. The International Journal
of High Performance Computing Applications, Volume 18, No. 1, Spring 2004, pp. 47-64.
Related Links
 Virtual Grid Application Development Software (VGrADS) Project
Parallel FFT Web Page
Gelato at the University of Houston (pdf)
A poster describing UH's Gelato-related projects as presented at the May 2004 Gelato Federation Meeting.
|
