Neptune is an Opteron Cluster in the Department of Civil Engineering at North Carolina State University and is the primary resource for R&D. Visit Neptune's website for more details. This page summarizes my contributions related to the cluster.
Planning and Deployment
-
Assessed our research group needs and identified specific hardware and software requirements.
-
Analyzed various hardware and software choices and recommended purchase decisions.
-
Deployed the hardware and performed initial configuration.
Programming Environment
-
Installed and configured MPICH and MPICH2, supporting both MPI and MPI-2 standards. Performed requisite configuration for using intra-node communication for processes residing on same node.
-
Installed AMD core math library (ACML) which incorporates BLAS, LAPACK and FFT routines.
Benchmarking and Optimization
-
Conducted extensive benchmarking using High performance Linpack(HPL).
-
Achieved a peak performance of 68.34 GFLOPS using 22 processors during preliminary benchmarking.
-
Further optimization involving HPL parameter tweaking for our specific architecture was done.
-
Achieved a peak performance of 76 GFLOPS with HPL(Linpack) while solving a solving a 70K x 70K dense matrix, corresponding to 78% of the theoretical peak.
-
Analyzed the performance of other benchmarks like High Performance Computing Challenge(HPCC) and NAS benchmarks.
Matlab and Distributed Computing
-
Installed the core Matlab environment and configured the Distributed Computing Toolbox(DCT).
-
Configured and deployed the Distributed Computing Engine(DCE) on the master and compute nodes on the cluster.
-
Conducted benchmarking of the Distributed computing features. Results are available at Neptune website.
Scheduler and Resource Manager
-
Achieved control over batch jobs and distributed compute nodes using Torque.
-
Configured Maui as the default job scheduler.
-
Configured and enforced basic Quality of Service(QoS) policies.
Source Configuration Management
Content Management System
|
