I’m a little tardy in hearing about this “48 core” problem in Linux - does anyone know if this is being addressed by any kernel programmers? I’m guessing it would appear in a hard-core distro (like OpenSuSE or RedHat) first. Most folks don’t care about this, I think…
Linux Scalability to 48 cores: the MIT analysis - Computing Performance | Google Groups
What’s got me confused is that technical folks have been clustering thousands of cores for years… why is this only being talked about now? I actually have the machine they mention in their research. I am not sure if this problem even shows up in MPICH/OpenMPI/etc. based numerical work - maybe it only shows up in highly multithreaded code. (I’m still trying to understand their paper)
A cluster of machines is not the same thing as a multicore CPU.
Of course - I was looking through their paper for references to MPI
Have any knowledge of the future? Since I have this machine, I’m interested in it, and which kernel(s) might be better. But, then, I might not care since I’m using this as a virtual cluster, running a bunch of MPICH processes rather than multithreaded processes. <noob face> I assumed GigE would be slower than inter-core communication. The downside is I can’t control the location (in memory or in core) of each MPICH process. I wonder if that means Gentoo would be better?