This subproject is one of many research subprojects utilizing theresources provided by a Center grant funded by NIH/NCRR. The subproject andinvestigator (PI) may have received primary funding from another NIH source,and thus could be represented in other CRISP entries. The institution listed isfor the Center, which is not necessarily the institution for the investigator.The objective of this project is to develop globus based methods for distributing certain types computational chemistry problems that require massive parallelization involving tens of thousands of processors. The best known example in this class is the 'folding@home' project, by V.J. Pande at Stanford University, for calculating transition states (long time scale dynamics) of large molecular systems. To achieve the kind of computing power needed for these types of problems, Pande and others, use volunteers around the world to donate spare CPU cycles from their desktop computers. An alternative to this approach is to distribute the computing over a grid of high performance computers. This type of approach has certain advantages. These include, but are not limited to, larger memory, larger disk space, a high speed Internet-2 connection (in the case of the Teragrid) between machines, and even higher speed connections between processors within a machine. In the approach we plan on using, the simulation of the physical system will take place on the Teragrid machines, while software running on a desktop computer will be used to coordinate processing on the Teragrid machines. The desktop machine will submit jobs to the Teragrid queues, act as a data repository, and coordinate load balancing on the Teragrid machines by keeping track of the processing that has been completed by the various processors on the Teragrid machines. The individual Teragrid processors will look to the desktop machine for guidance on where to send output data, what data to process next, and, if necessary, where to get the next data set for processing. Message passing within a machine will be based on one-sided (non-blocking, remote memory access) mpi calls. Since not all the machines on the Teragrid support this feature, we will be restricted to a subset of the Teragrid. Despite this disadvantage, the use of remote memory access will greatly simplify load balancing, and it is very likely that in the future this feature will become more commonplace. Communication between machines will be achieved using globus_xio with the TCP, TCP-XM (a multi-cast TCP emulator), and GridFTP drivers. The ultimate goal of this project is to determine the efficiency and effectiveness of this type of parallelization on various types of computational chemistry problems. The initial test case will probably be a distributed quantum Monte Carlo run. Later test cases will involve algorithms that require a distributed octree data structure.
Showing the most recent 10 out of 292 publications
