Can anyone tell me what type of effect the location of compiliers on a cluster has on Software compiled with the compiler has? Im thinking about two cases. First, a complier is shared across all nodes by NFS. Second, a complier is installed on each node. Obvisiously the second choice is optimal, however how much of a difference chould it make? Chould it be considered negiliable on a small (say 12 node) cluster?
High-availability or failover clusters are especially useful for
mission-critical databases, services and files for business
applications.. They are based on several different redundant servers or nodes that replicate data, programs and server functions so that when
components fail, one of the nodes can resume service without any
noticeable interruption in service..
It's fairly typical to have a very small number of nodes set up for compilation, and mount the shared libraries on NFS, if the local practice is to build against shared libraries. For Intel compilers, you can use the redistributable libraries if you choose to install them on all nodes, regardless of how you license the compiler (node-locked or by floating license server). Commercial applications often are static linked against the compiler libraries, and dynamic linked against the MPI libraries and the libraries which the OS provides, expecting those libraries to be installed on each node, while the application may be installed on NFS file share.