Often in computational astrophysics, we have initial conditions given to us by a one dimensional (perhaps stellar evolution) code. These codes are usually - for convenience - formulated in a Lagrangian coordinate system, where each coordinate (e.g. mass) moves along with the fluid. Multidimensional hydrodynamics codes typically use an Eulerian coordinate system in which the coordinates (e.g. position) are fixed and the fluid moves through the them. Mapping a unequally-spaced Lagrangian model into an Eulerian code with a uniform mesh while preserving hydrostatic equilibrium can sometimes be challenging. Here I describe a variant of the procedure we often use in the CASTRO and MAESTRO codes; this is based on the paper by Mike Zingale et. al.
The goal is to set up svnserve for svn repository access using the svn network protocol on a server running RHEL or one of the clones (CentOS and Scientific Linux). We'll configure svnserve to use the Cyrus Simple Authentication and Security Layer (SASL) library for authentication and encryption if the svn client is built with SASL library (e.g., those on most of the popular Linux distros); and to fall back on using the built-in CRAM-MD5 mechanism for authentication (but no encryption) if the client is not built with SASL library (e.g., the stock svn client on Mac OS X).
Hopper is the flagship supercomputer at NERSC. It is a Cray XE6 system, with a peak performance of 1.28 Petaflops/sec. It is, to the best of my knowledge, also the only Cray supercomputer that has Intel compiler installed. Intel compiler is my favorite. It typically produces binaries that are more efficient than those produced by its competitors, particularly a lot more so than the GNU compiler. Even on AMD processors! However, the performance of Intel compiler on Hopper was puzzling. As observed by Dr. Andy Nonaka in 2011,
Intel beats both GNU and PGI in serial and pure MPI, but once you turn on threads (MPI + OpenMP), both PGI and GNU are more than twice as fast.
