I'm trying to understand the behavior of the Linux pdflush daemon when:

• The page cache is initially pretty much empty
• There is a large amount of free memory
• The system starts undergoing heavy write load

My understanding right now is that the vm.dirty_ratio and vm.dirty_background_ratio that control page cache flushing behavior are with respect to the present size of the page cache, which means that my writes will flush earlier than they would if the page cache was pre-populated (even with dummy data from some random file), and thus throughput will be lower. Is this accurate?