From: Andi Kleen <ak@linux.intel.com> We had a case where a 4 socket system spent >80% of its total CPU time contending on the global urandom nonbloc
↧