Back to Volume
Paper: Morphological Filtering of Infrared Cirrus Emission on Parallel and Distributed Computers
Volume: 61, Astronomical Data Analysis Software and Systems III
Page: 308
Authors: Pedelty, J. A.; Appleton, P. N.; Charmandaris, V.; Basart, J. P.
Abstract: We have implemented on a variety of parallel and distributed computers the infrared cirrus filter developed by Appleton, Siqueira, and Basart (A.J., in press, 1993, or see the Proceedings of ADASS 1, pg. 283). The filter uses the techniques of mathematical morphology (a.k.a. morphological image processing). The key feature of morphological image processing is its ability to discriminate features based on shape (hence the name). The shape to be probed is contained in the structuring element. Initial results using a Gaussian structuring element indicate an ~15x reduction in the average level of cirrus emission in the 100microns IRAS image of the M81/M82 field. The filter is computationally expensive, and so we were motivated to explore the use of parallel and distributed computing to reduce the filtering time. The parallel processing platforms we have used are the MasPar MP-1 and MP-2, the Thinking Machines Corporation CM-5, the Intel Paragon, and the Kendall Square Research KSR-1. We have also made some initial tests using homogeneous and heterogeneous networks of distributed workstations linked via the Parallel Virtual Machine (PVM) environment. We report the performance attained with each system, and compare to that obtained with traditional vector processors such as a Cray Y-MP and Convex C3240, workstations such as a DECstation 5000, and the new generation super-workstations made by DEC and HP. For example, the MasPar MP-2 with 16K processors is 350x faster on this problem than a DECstation 5000/240, and is 10x the performance of a Y-MP processor. We discuss in more detail the implementation of the filter on the MasPar computer. The MasPar is a 2-d mesh of up to 16K processors tightly coupled to a DECstation front-end computer which runs the Ultrix operating system. We thus can run the filter on the MasPar just as we would on a Unix workstation. The filter is written in MasPar Fortran (MPF), a language which has many features from the Fortran 90 standard. We have used the FITSIO routines developed by W. Pence at NASA/GSFC to provide a direct FITS I/O capability to the MasPar filter. The dramatically faster performance available from the MasPar has made it much easier to experiment with the parameters of the cirrus filter. We report the initial results of filter tests made using structuring elements with a variety of shapes and amplitudes. This work is funded by the NASA High Performance Computing and Communications Program (HPCCP).
Back to Volume