9500 Gilman Dr. #0505
La Jolla, CA 92093-0505
glock at sdsc dot edu
As of July 11, 2014, I no longer work at the San Diego Supercomputer Center. This site remains here for reference, but please be aware that it is no longer being updated. I am maintaining a permanent, personal mirror in preparation of my SDSC site deactivating.
I used to be a user services consultant at the San Diego Supercomputer Center and provided all manner of support to users of its high performance computing resources provided through NSF's XSEDE program. My scientific background is materials science, and I have specific expertise in silica, water-mineral interfaces, and molecular dynamics simulations. Supercomputing was a hobby-turned-profession for me, and although my interests in the field are broad, they all converge on understanding the demands for and uses of cyberinfrastructure to make supercomputing less obtuse and more accessible to researchers and the public. More information is available on my LinkedIn profile.
This site contains various notes I had taken that may be relevant to a broad audience. I should emphasize that the contents of this site do not reflect the advice or endorsement of the San Diego Supercomputer Center, and all files, guides, and notes are my own and are wholly unsupported by SDSC. If you are looking for official support or help, see the SDSC User Guides or contact firstname.lastname@example.org.
Technical/High Performance Computing
- Common Problems on SDSC's Supercomputers
- Installing software at SDSC without root (incomplete)
- Configuring SSH Keys on Windows
- Gaussian: Introduction to Gaussian on Linux
- DDT: Getting started with DDT on SDSC Gordon
- Topics in Data-Intensive Computing (intro)
- Hadoop: Conceptual Overview of Map/Reduce and Hadoop
- Hadoop: Running Hadoop Clusters on Gordon
- Hadoop: Writing Hadoop Applications in Python with Hadoop Streaming
- Hadoop: Parsing VCF Files with Hadoop Streaming (incomplete)
- R: Using R on SDSC Gordon and Trestles
- R: Parallel Options for R
- R: Parallel R using Hadoop (not even started!)
Application Install Guides
- Building ReaxFF support for LAMMPS on SDSC Gordon
- Building Thread Building Blocks on SDSC Trestles
- Building IPM 0.983 for lightweight MPI profiling
- Notes on map/reduce implementations (perpetually incomplete)
- Managing processor affinity in Linux
- Setting up a Quick MPI Cluster on Amazon EC2
- Benchmarks for various processors and architectures
- Benchmarks for MPI on Amazon EC2, Blue Gene/P, and Myrinet 10G interconnects
- Benchmarks for MPI on Amazon EC's SR-IOV-enabled C3 instances vs. native 10 Gbe
- Benchmarks for MPI and Applications InfiniBand virtualized with SR-IOV
- Benchmarks and tuning for dual-rail QDR vs FDR InfiniBand
- Benchmarks for TCP vs. RDMA over InfiniBand
- List of UltraSPARC-based workstations
- Computer power consumption data
- SCSI quick reference with pictures
Many of these files are now hosted on my GitHub repository.
A nicer interface to Torque/PBS for SDSC Gordon and Trestles
Finds processor sockets with no high-load processes. Good for scheduling NUMA-intolerant SMP jobs on a single socket in shared environments.
- LAMMPS files
- Makefile for SDSC Gordon - Build against our Intel/MVAPICH2/MKL
- Makefile for SDSC Gordon - Build against our Intel/MVAPICH2/FFTW3
- Makefile for SDSC Trestles - Build against our PGI/MVAPICH2/FFTW3
- Makefile for Rutgers Excalibur (Blue Gene/P) - Build against IBM XL compilers, IBM's MPI stack, and IBM's ESSL FFT libraries with the FFTW3 interface
- Various useful notes on R by Jonathan Olmsted at Rochester. In particular his notes on running R on clusters was particularly helpful for me when scaling up from workstation to supercomputer.
- Gordon's Architecture
- Torus hop distance calculator for Gordon
- Gordon system status
- Trestles system status