ChemIO

High-Performance I/O for Computational Chemistry Applications

The ChemIO project is a joint effort of Argonne National Laboratory and Pacific Northwest National Laboratory. It is affiliated with a DOE Grand Challenge project developing Massively Parallel Methods for Computational Chemistry, with the multi-agency Scalable I/O Project, and with the EMSL. It has three main goals:

To facilitate the development of portable, high-performance computational chemistry codes, by defining a standard I/O API that meets chemistry requirements, and providing high-performance implementations of this API on different high-performance computers;

To contribute to the understanding of requirements for standard APIs for high-performance I/O, the performance that can be achieved for these APIs on various high-performance computer systems, and the techniques required to optimize performance on various systems; and

To investigate how the successful Global Array programming model can be extended to secondary storage.

Computational chemistry is an extremely interesting discipline from a high-performance I/O standpoint. Certain computational chemistry codes perform well on scalable parallel computers. They often operate on very large data sets and display irregular access patterns. Furthermore, many state-of-the-art computational chemistry codes have been engineered to perform repeated recomputations so that they can run in-memory on vector computers; it seems likely that I/O-based ("out-of-core") solutions will allow significantly better performance on scalable parallel computers.

A paper accepted for publication in a special issue of The Int. Journal of Supercomputer Applications and High Performance Computing provides further details on the ChemIO project.

Progress, Software, and Users

So far we have produced a design and implementation for a chemistry I/O API, ChemIO, comprising three components:

Disk Resident Arrays(DRA), an extension of global arrays to secondary storage;

Exclusive Access Files (EAF) per-processor private files; and

Shared Files(SF): files to which multiple processors can read and write independently.

All the components have asynchronous API to allow the applications overlap expensive I/O with computations.
The ChemIO is implemented in a a modular fashion. The user level librarries (DRA, EAF, SF) are layered upon ELIO (ELementary I/O) device library that provides an portable interface to different file systems.
On IBM SP, Shared Files use Distant I/O library in addition to ELIO.

ChemIO code is in use by several application groups. For example:

A group at Argonne are using ChemIO's Exclusive Access Files to build a high-performance non-direct (out-of-core) SCF code (see also their benchmark results)

RI-SCF and RI-MP2 programs being developed at PNNL and Syracuse use DRA

semi-direct SCF implemented in NWChem (PNNL) uses EAF

For More Information ...

Contact Ian Foster (foster@mcs.anl.gov) or Jarek Nieplocha (j_nieplocha@pnl.gov).

If you are interested in using this software, please contact the developers.