“High-Performance Computing,” or HPC, is the application of “compute nodes” to computational problems that are either too large for standard computers or would take too long individually. HPC typically consists of a system manager server (SMS, also know as login node or master node, or all combined) and compute nodes. HPC designs may differ but frequently offer high speed networks, large home directories, scratch space, archive space and a job scheduler. A provision application is used to (re)image the compute nodes when needed.
Please answer any General and/or Technical questions that apply to you and/or your group. The intent of this survey is to properly size and identify components our computational environment would look like. The time horizon is roughly “within the next three years”.
Q: Regarding your primary computational needs, identify your discipline and/or department.
Q: In two sentences identify your primary research area/project.
Q: If you do not currently use HPC, do you anticipate using it in within next three years?
Q: If you currently use HPC (local or remote), please give brief description (ie I use Jetstream/Stampede/…my own cluster).
Q: Does your Department anticipate expanding/initiating computational activities in the curriculum, recruiting, or current research projects?
Q: Do you currently use or anticipate the need for large storage capacity?
Q: Do you currently use or anticipate the need for high interprocess communications, frequently referred to as running parallel programs across multiple compute nodes?
Q: Are your typical compute intensive applications CPU bound (in memory computations only), IO bound (perform lots of reads and writes to storage), or both?
Q: Will be you be anticipating, or are you currently using, GPU computing?
Q: Most applications will be floating point operations, however do you use integer based computing?
Q: How would you like to participate in ongoing developments of our HPC (ie email/list, meetings, named contact …)?
Q: Would you support some level of periodic “monetary contributions” based on your groups' cpu usage to build a budget for future HPC maintenance.
Q: What Linux distributions are you using or familiar with (ie CentOS, Redhat, Suse, Ubuntu …)?
Q: What Scheduler are you using or familiar with (ie SGE, Torque, BPS, Openlava …)?
Q: What commercial grade compiler are you using or anticipate needing (ie Intel's icc/ifort, PGI's pgicc/pgfortran …)
Q: What commercial software do you use or anticipate using (ie SAS, Stata, Matlab, Mathematica, IDL/Envi …)? How many concurrent licenses for your group?
Q: What open source software do you use or anticipate using (ie Gromas, Amber, Lammps, R …)
Q: For the two questions above: how important is it to you that your job can restart where it left off when a crash happens (versus starting over)?
Q: Do you run parallel programs using MPI across multiple compute nodes, or anticipate doing so? Which MPI flavor (OpenMPI, MVApich, OpenMP …)?
Q: Do you run forked programs (threads) confined to a single compute node? Which application (ie Gaussian, Autodock …)?
Q: For the two question above: how many cpu cores or threads per job typically, typical total memory requirements per cpu core or thread (or total per job).
Q: For serial jobs (one program requires one cpu core); typical total memory requirement per job.
Q: For all type of jobs listed above; how many concurrent jobs would be running, for how long per individual job, in a typical research month or quarter.
Q: Estimate the size of your home directory needs for actively used files assuming archive space is available (ie 100G, 1T …).
Q: Estimate your archive space needs at the end of three years (ie 1T, 5T …).
Q: Estimate the need for large scratch space requirements (ie jobs that write 100-500G of content per job).
Q: For home and scratch space areas, do you require a parallel, clustered file system or are simple NFS mounts sufficient?
A: Give us any feedback we did not ask for or forgot to ask