Lessons Learnt from using .NET for Distributed Finite Element Analysis

The paper considers the use of .NET as a framework for use in distributed scientific computing on desktop computers. Execution speed tests are carried out for a range of problems. The range covered direct and iterative solution, and domain decomposition methods. Tests were carried out on dual and single processor machines, and on clusters of single processor machines. The code used was written in C++ and C#. The C# code was written in both safe and unsafe versions. Some tests were also carried out using LaPack++, and MPI was used to implement distributed and parallel solution with C++. The speed differential was found to be machine dependent, but in many cases the unsafe C# code was less than 10% slower than C++ code. Using .NET for distributed computing did not incur any noticeable overhead for direct solvers, but did do so for implementing iterative solvers on a cluster of computers.