The comment for balance_dirty_pages_ratelimited explains rate limiting is necessary because "On really big machines, get_writeback_state is expensive". This comment is stale, since get_writeback_state has not existed for more than 10 years. I gather, however, that its expense arose from aggregating vmstats over all cpus. These days, global vmstats are simply a memory load. Is there a modern reason for rate limiting balance_dirty_pages? Thanks, Benjamin
