Post #2022847

Overconfident Guy on the Internet: > "there should be no dependence on swap existing whatsoever as long as you have more than enough memory for the entire workload." Dear Confident: we regret to inform you that dead pages in low-to-no utilization daemons are not 'workload.' Furthermore, we refer you to the concept of translation lookaside buffer crowding. Perhaps you might find illuminating investigating the ways in which the TLB repopulation cost of context switching increases latency? Yours truly, Actual Engineering p.s. Rule of thumb for overconfident internet guys: cache repopulations suck even harder if the cache is full of dumpster fire pages.