Equation Solution  
    High Performance by Design
Navigation Menu
Home  ⇒  Parallel Computing  ⇒  LAIPE  ⇒  Java LAIPE  ⇓

Java LAIPE: parallel direct solvers

        Java LAIPE is a Java version of LAIPE. Similar to LAIPE, Java LAIPE delivers linear speedup. For example, the following timing result is implemented on a UNISYS Clearpath quad 200MHZ Pentium Pro.


        This example applies a parallel skyline solver for a sparse asymmetric system of [A]{X}={B} of order (11,111x11,111), a total of 3,262,339 non-zero fillins. The timing result and linear speedup are as:

Number of CPUs Elapsed Time (sec) Speedup Efficiency (%)
1 21.578 1.00 100.0
2 10.890 1.98 99.0
3 7.313 2.95 98.3
4 5.547 3.89 97.3


        This example applies a parallel band solver to decompose a sparse, symmetric, and positive definite matrix of order (10,000x10,000) with a constant bandwidth 600 into triangular matrices. Timing result, speedup and efficiency are as:
Number of CPUs Elapsed Time (sec) Speedup Efficiency (%)
1 123.66 1.00 100.0
2 62.81 1.97 98.5
3 42.78 2.89 96.2
4 33.17 3.73 93.2
With the nature of asynchronous parallelism, LAIPE, no matter in Fortran or Java, always delivers a highly efficient speedup. Most parallel solvers (or parallel programs) never disclose speedup. LAIPE is different. LAIPE can yield a highly efficient performance, and always discloses speedup.

        Java LAIPE is in Java MTASK which is a Java version of MTASK. Java LAIPE is directly converted from fortran version of LAIPE. For information, please refer to LAIPE. Java LAIPE provides powerful parallel direct solvers.