External:ComputationalScience
From Wiki
This page is an information-collection point for computational scientists on campus. It was started prior to a meeting of several groups called by IAM on December 11th, 2008 to help figure out what is needed to support scientists' computational needs on campus. Everyone involved should feel free to create an account on this Wiki and add things in both before and after the meeting. Please also add to the Assets section if you know if things that others might want to make use of.
Wiki feature: If more than one person edits at the same time, the first person's changes will be overwritten, but they will be stored in the history of the page, so can be resurrected and re-inserted.
Contents |
Needs
Trust or local control
The demonstrated ability of those with control of the systems to be responsive and to not do things that are destructive to the researcher's environments and data or else the researchers must own the data and environment themselves.
Warning before things are changed, and the need to ask permission before things are changed in the environment that adversely affect the scientists' workflows.
OS/grid support
People to help small clusters (maybe 16-100 machines) that are already in departments.
Responsive systems administrator who can set up virtual machines and keep the operating systems up and running. Sitting down the hall from each research group and evaluated by them.
Provide a development environment that is identical on workstations, small clusters, large clusters.
- a debugger on all platforms (e.g. Totalview)
- a profiler on all platforms
A software environment that is stable on a 3-year cycle (libraries, compilers). Security can be ensured by isolation. Perhaps running software in a virtual machine.
Applications/Libraries
Person to support the installation and maintenance of software packages needed by groups around campus.
Money to purchase software tools used by researchers on campus.
Training
Information on what is available, and updates on what will be changing before it happens.
Porting/parallelization support
Moving applications from laptop-scale up to small and then large cluster scales. In some instances, this is just showing how to run a bunch of single jobs independently. At other times, it will be help making applications parallel.
Support development of applications at UNC.
Mechanism for pooling resources
How to put support for system support and perhaps hardware into grant proposals and pool the resources. Needs to be a recharge center or something like that that will pass muster at the funding agencies.
Hardware
Data Visualization support
Assets
Hardware
Campus Research Computing (ITS): ITS operates shared computational facilities including compute clusters, large-scale file systems, mass-storage systems, and specialized software for use by campus research projects. Facilities include
- Topsail: 520 node (4,160 CPU) cluster with 12GB memory per node and Infiniband 4x SDR interconnect, available for parallel campus jobs [1]. Eighty percent of the cycles on this cluster will be allocated using an allocation process [ http://www.renci.org/unc/computing/resources/allocations.php ] requiring submission of a proposal detailing research to be done, computational requirements, number of CPU hours needed, and including demonstration of the scalability of code to a large number of processors (if large parallel resources are requested). At most times, up to 20% of the system's cycles may be available for other work on a first-come, first-served basis.
- Emerald: ~150 node (~350 CPU) heterogeneous cluster with 2 - 128 GB memory per node.
- Cedar/Cypress: 128 CPU, 512 GB shared memory SGI Altix 3700.
BASS (Russell Taylor, CS): NIH-sponsored facility for researchers doing biomedical analysis and simulation. The primary user group for the BASS supercomputer [2] are those associated with CISMM, BRIC, IAM, the Virtual Lung Project, MBRL, and RENCI. An advisory board with members from these groups sets priorities on the machine.
- BASS: ~60 node (452 CPU) heterogeneous cluster with 8 - 128 GB memory per node and Infiniband interconnect.
- BASS - GPU: 45 nodes of BASS have 2 or more attached GPUs. Researchers developing GPU-accelerated parallel codes using CUDA are in the primary user group for the extra GPUs donated by NVIDIA for the BASS supercomputer [3]. Contact the PI (Russ Taylor) for scheduling; these projects do not need to be NIH funded.
RENCI: Renci has significant computational resources and computing expertise to support collaborations on computational projects statewide and nationally.
OS/grid support
Applications
Training
LearnIT: ITS provides a variety of training courses on campus on a wide variety of topics [4]. A current schedule of courses, including Linux and Computational Chemistry, is available at [5]. Recent past courses have included using MPI to parallelize programs, an introduction to Topsail, and Introduction to Scientific Computing.
Porting/parallelization support
BASS: NIH-sponsored campus researchers doing biomedical analysis and simulation can call on the half-time support person funded out of UNC match for the BASS supercomputer [6].
Potential Assets
This section is for brainstorming about potential things that might be helpful and how the needs might be met. It is not binding on anybody, but provides a place for people to toss in ideas and possible help. A pressing need is to provide input for the upcoming Capital Campaign. One suggestion is that we create two scenarios: 1) an 'immediate needs' list of equipment, facilities and support, perhaps in the $5M range. 2) an expansive list of equipment, facilities and support to take UNC-CH 'to the next level'. This might be significant fractions of the $250M 'units' of the eventual $4B ask.
OS/grid support
CompSci Recharge Center: The Computer-Science department has a staff that handles operating-system, networking, and application support for the CS community. This has been extended to particular collaborators across campus who are working closely with CS; such researchers pay a fraction of the full-time CS fee to the recharge center in return for particular services. This could be a mechanism to extend support more broadly.
Training
Computers in X: CS faculty have been talking about the providing themed introductory programming courses, with flavors for different on-campus groups. Computers in Chemistry, Computers in Biology, and so forth might be possible such offerings. No concrete plans yet, so it is a good time to provide feedback on which groups would most benefit from this and how we could provide the needed resources to teach and TA the courses.
Visualization in the Sciences: Russ Taylor teaches a cross-listed course in CS, Physics & Astronomy, and CASE on tools and techniques for the visualization of scientific data sets, including 2D and 3D scalar, vector, and tensor simulation data. This course is taught each spring. Interested scientists are welcome to audit or cherry-pick lectures that are the best fit to their data sets and questions. The course includes a semester-long project section, during which teams of students design and construct prototype visualization tools for specific problems. Scientists across campus are encouraged to submit data-visualization challenges as potential projects.
