more accurate memory limit calculation for anon-r-seq cases Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>