Future Technologies News Archive

2008-03-31 -- FT Demonstrates High Performance InfiniBand on Wide-Area Networks. FT group members Weikuan Yu and Jeffrey Vetter, along with Nageswara Rao (also of Computer Science and Mathematics at ORNL), will be presenting their recent analysis results on high performance InfiniBand on wide-area networks in the 2008 IEEE Network, Architecture and Storage conference in Chongqing, China. Using the OC192 SONET connections from UltraScience Net, they have demonstrated that a network-level bandwidth of 7.4Gbps and an MPI-level bandwidth of 758MB/sec are achievable across two InfiniBand clusters that are 8600 miles apart. Part of results will also be presented in the OpenFabrics Conference in Sonoma, California.

2008-02-29 -- FT IO team releases its optimized MPI-IO driver for Lustre as a part of MVAPICH. The IO working team of the Future Technologies group has released its optimized MPI-IO driver for Lustre as part of MVAPICH-1.0, the popular MPI package from the Ohio State University, Network-Based Computing Laboratory. This release enables Lustre stripe-aligned high performance parallel I/O for scientific benchmarks and applications such as NAS BT-IO and FLASH I/O applications. Further info on code download, installation, and performance is available on MVAPICH website. The initial package was developed for use on the Cray XT system at ORNL.

2008-02-28 -- FT Reports on Low-Overhead Virtualization and Light-Weight Migration of HPC Parallel I/O using Xen and PVFS. FT group members Weikuan Yu and Jeffrey S. Vetter will be presenting their paper "A parallel I/O perspective for Xen-Based HPC" at the Eighth IEEE International Symposium on Cluster Computing and the Grid, in Lyon, France. The paper presents an analysis of HPC parallel I/O in a virtualized environment. Using a PVFS-based virtual cluster, the team has analyzed the overheads and migration costs for parallel I/O programs in a Xen-based virtual machine cluster over two different networking protocols: TCP-based Gigabit Ethernet and VMM-bypass InfiniBand. The authors demonstrated that Xen-based HPC environment can support high performance parallel I/O with negligible overheads, and the migration of parallel I/O programs will add little cost to the total execution time.

2008-02-28 -- FT Investigation on Characterization and Optimization of Large-Scale Parallel I/O on Jaguar. FT group members Weikuan Yu and Jeffrey S. Vetter, along with Sarp Oral of the NCCS Technology Integration group, will be presenting their paper "Performance Characterization and Optimization of Parallel I/O on the Cray XT" at the 22nd IEEE International Parallel and Distributed Processing Symposium in Miami, Florida. The paper presents an extensive characterization, tuning, and optimization of parallel I/O on the Cray XT supercomputer at Oak Ridge National Laboratory. This characterization covers both data- and metadata-intensive I/O patterns. The authors shed light on the effectiveness of several parallel I/O techniques, such as data sieving and two-phase collective I/O, on the Cray XT. Moreover, the authors have demonstrated that it is possible, and often prudent, to improve the I/O performance of scientific benchmarks and applications by carefully optimizing their application's I/O operations. For example, they show that the I/O performance of the S3D combustion application can be improved at large scale by tuning the I/O system to avoid a bandwidth degradation of 49% with 8192 processes when compared to 4096 processes.

2008-02-27 -- FT investigation reveals popular scientific benchmarks do not accurately represent TLB behaviors of real applications. FT group members Collin McCurdy and Jeff Vetter, along with Alan Cox of Rice University, will be presenting their paper "Investigating the TLB Behavior of High-end Scientific Applications on Commodity Microprocessors," in Austin, TX this April at the 2008 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS’08). The paper is the culmination of work, undertaken as part of the PetaSSI FastOS project, seeking to understand the TLB behavior of scientific applications. The analysis shows that two benchmark suites that are understood to represent scientific application behavior (SPEC CPU and HPCC) are not representative of the TLB behavior of important full scale applications. Furthermore, the paper demonstrates that false conclusions drawn from benchmark TLB performance can have significant ramifications for application performance.

2007-11-29 -- Junqing Sun wins SC07 ACM Student Research Poster Competition. As part of Future Technologies Reconfigurable Computing (FPGA) research initiative, Junqing Sun, in collaboration with Olaf Storaasli (ORNL) and Greg Peterson (UT) has completed his PhD thesis, entitled High Performance Reconfigurable Computing for Linear Algebra: Design and Performance Analysis from the Department of Electrical and Computer Engineering at the University of Tennessee. A key thesis focus, Obtaining High Performance via Lower-Precision FPGA Floating Point Units was presented at Supercomputing 2007 (and won 1st place in the ACM Student Research Poster Competition). Obtaining double precision accuracy via iterative refinement of single precision, an idea heralded by Jack Dongarra (also on Junqing's Ph.D. Committee), has now, for the first time, been shown to significantly reduce computation time on FPGAs while still preserving double precision accuracy.

2007-11-28 -- ORNL FT Group to give PPoPP'08 tutorial on Extreme-Scale Parallel I/O. The ORNL Future Technologies group with personnel from the ORNL National Center for Computational Sciences are to give a tutorial on parallel I/O for extreme-scale platforms such as the Cray XT, IBM Blue Gene, and ultra-scale Linux clusters. We will share our experiences and lessons learned from characterizing and optimizing I/O on these platforms, and discuss directions and issues that need to be addressed for I/O at the unprecedented scale needed for petaflop and even exaflop systems. Join us on February 23, 2008 in Salt Lake City for this enlightening, in-depth discussion.

2007-11-09 -- FT updates its MPI-IO driver for Lustre. To meet the increasing interest of the community, the IO working group at Future Technologies Group has posted an updated version (v03) of its MPI-IO driver for Lustre. This version is made compliant with the latest MPICH2-1.0.6p1 release, along with other feature enhancements.

2007-10-14 -- ORNL FT Group coordinating SC07's Disruptive Technologies Exhibits. This year, the ORNL Future Tech group is coordinating the SC07 Disruptive Technologies activity, which includes Exhibits and Panels. Winners for the DT Exhibits showcase have just been announced; they include quantum computing, many-core chip system software, cmos photonics, NV-RAM based on carbon nanotubes, and optical printed circuit board technology. SC07 attendees can visit these exhibits in Exhibit Hall 1B of the Reno-Sparks Convention Center from Monday, Nov. 12 to Thursday, Nov. 15

2007-08-28 -- Balancing Productivity and Performance on the Cell Broadband Engine. The Cell Broadband Engine is a heterogeneous multi-core processor capable of very high performance; however, users must explicitly manage data movement, scheduling, and synchronization. While these attributes provide some of the cell processor's greatest performance strengths, they also form its greatest weaknesses in terms of developer productivity, code portability, and initial performance efficiencies. At the IEEE Cluster 2007 conference in September, a paper by Sadaf R. Alam, Jeremy S. Meredith, and Jeffrey S. Vetter will be presented which evaluates how to achieve high performance on the Cell processor while maintaining high productivity across a diverse set of application workloads.

2007-07-14 -- FT posts an initial MPI-IO driver for Lustre. The IO working group at Future Technologies Group is mounting up their efforts in enhancing parallel IO for extreme-scale computing. For culturing more research and development on parallel IO issues related to Lustre-based supercomputers, FT posted an initial MPI-IO driver. The long-term objective is to provide a high-performance, open-source, yet unified code base that works for Cray XT and other platforms. This is a joint effort with our collaborators, such as Technology Integration Group at National Center for Computation Sciences, ORNL.

2007-06-22 -- FT Group welcomes Summer Interns. Summer interns from around the nation joined ORNL's Future Technologies group this month. The interns will examine a number of technical research topics that include emerging architectures and HPC system software. The interns are Randy Hassell from Old Dominion University, Nicholas Murphy from University of Washington, Viraj Paropkari from Georgia Tech, and Srinivas Sridharan from University of Notre Dame.