Applications
To support the ESS Grand Challenge Project teams, to achieve the HPCC project milestones and to make their codes available to the National HPCC Software Exchange. We will work with the JPL ocean modeling group leaded by Dr. Yi Chao, make their code running efficiently on distributed or shared memory systems, and provide necessary technical assistance.
In order to achieve HPCC project milestones for the ocean modeling, implementing a well-designed parallel ocean code and improving the computational efficiency of the model (OGCM) is essential. There are many challenges to design an efficient ocean modeling code on parallel systems, such as how to design an interface with other Earth System Model elements (Atmosphere Model AGCM and Tracers Chemistry Model ACM). This can be done by implementing a data broker for coupling individual codes.
Ocean modeling is one of the grand challenges of computational science, and three-dimensional time-dependent models require a large amount of memory and processing time to run realistic simulations. The HPCC program provides each grand challenge team certain amount of computing time on the HPCC testbeds, but it is still a good idea to have one's own supercomputer for daily testing and coding. With the growing power and shrinking cost of personal computers (PCs), the availability of Gigabit ethernet interconnect, and public domain software packages, it is now possible to combine them to build desktop parallel computers (named Beowulf or PC clusters) at a fraction of what it would cost to buy systems of comparable power from supercomputer companies, such as Cray Research Inc./Silicon Graphics, Inc. or Hewlett Packard. Recently, "do-it-yourself" supercomputers, such as Beowulf system built from ordinary personal computer components, become powerful tools for computational science community. To develop a prototype desktop parallel computer for ocean modeling will have significant implications in the scientific computing, and it will also help the ocean research team to achieve the HPCC project milestones on time.
On the software side, a data broker has been successfully implemented by several team members. The coupled AGCM/OGCM/ACM system runs 4.7 times faster than the coupled AGCM/OGCM system used in performance milestone 1. The milestone was achieved on 787 nodes of the NASA GSFC CRAY T3E-600 machine. 397 nodes were dedicated to the coupled AGCM/ACM system and 390 nodes to the OGCM. These performance numbers were measured by running the models for 10 simulated days.
On the hardware side, we have been successfully built our Beowulf system (Fig. 1) with the support from the JPL Director's Research and Development Fund (Y. Chao, P. Wang, and B. N. Cheng). This system consists of 16 Intel Pentium II PC's running at 300 mhz, interconnected by a 100 mbs fast ethernet network, with a total price of about $32K. Now it has been used by the ocean group as a daily working system.
Figure 1. The Ocean modeling Beowulf cluster-Jaws.
The implementation of the data broker allows individual codes to execute on a parallel system as a single code that gives scientist a great potential to attack the global climate problem by using advanced computing systems. The Beowulf cluster brings significant computing time for the ocean group in JPL. Based on our experience of building the cluster and using it as our daily system, we conclude that the Beowulf class supercomputers are well suited for ocean modeling purposes, and with the current trends in PC pricing, seem destined only to grow in suitability. The attractive price-to-performance ratio means such machine is likely to be around for research and many other non time-critical applications.
Currently, the ocean code runs well on various parallel system. The updated ocean code has significant speed up through the optimization work on each system. Through a data broker, it can be easily coupled with other Earth system codes. The Beowulf cluster in the JPL ocean group is in a fairly stable condition. It has been extensively used by the ocean group members since it was born in the beginning of last year.
Ping Wang
Jet Propulsion Laboratory
wangp@rockymt.jpl.nasa.gov
818-393-1941
References
[1] P. Wang, B.N. Cheng, and Yi Chao , " Climate Ocean Modeling", in High Performance Cluster Computing , Vol. 2, 526-539, Prentice Hall Publisher, 1999.
[2] B.N. Cheng, P. Wang, Y. Chao, and M. Bondarenko, "Climate Ocean Modeling on a Beowulf Class System", Proc. of International Conference on Parallel and Distributed Processing Techniques and Applications, Las Vegas, 1999.