My scenario is a 1TB Intel workstation with Fedora 20 with a single process using almost all memory (and most memory allocated by the process comes from two almost-half-terabyte allocations; then there is a very small number of smaller allocations). I would like to exploit as much as possible huge pages to get a faster memory access. Unfortunately, the documentation is very scarce. Actually, most documentation/questions are about how to disable huge pages.
So, in order:
- Which would be the right parameters for the kernel command line, or in sysctl.conf, to enable huge pages and transparent huge pages?
- How can I obtain a hugepage allocation? Is it sufficient that the amount of RAM required is an exact multiple of the hugepage size, as it seems from the documentation of transparent hugepages?
- Will these settings affect in any way processes who do not perform large allocations?
- How can I check that I'm actually performing hugepage allocation (I guess hugeadm --explain is sufficient, but I'd love to get a confirmation)?