If your large dataset leads to a single huge memory allocation, the kernel will fail the allocation even though overcommit is enabled. Early versions of libvpx

If your large dataset leads to a single huge memory allocation, the kernel will fail the allocation even though overcommit is enabled. Early versions of libvpx had a similar issue, where it allocated a huge amount of memory at once (actually, told the dynamic linker to do so) which failed on low-memory machines (https://bugzilla.redhat.com/show_bug.cgi?id=600663 https://bugzilla.redhat.com/show_bug.cgi?id=600663 "libvpx.so.0 fails to load on low-memory machines due to ridiculously oversized .bss").