Neither OpenMOSIX nor OpenSSI are normally used for HPC clusters because they scale not very well to a high number of nodes and have too high overhead. Especially SSI is more aimed at the HA space, which has completely differnet requirements.

Three things are quite clear: 1. there shoul be *some* overhead in SSI. Do you have any good links concerning this overhead and what is the progress with this overhead problem? 2. handmade parallelisation should always show better performance. Here the situation is quite similar to "assembler<->high level language". Indeed, one could easy agree that high level language bring *some* overhead. However majority of programmers do not use assembler for general purpose programming anymore. So, it would be interesting to know at which time-point are currently the "high-level-parallelisation tools" like OpenSSI and OpenMosix. 3. shared memory is a problem, of course, but not all projects use the shared memory. (e.g. my software is potentially good for process migration because of no shared memory usage) Thank you for your remark.