On 5/21/21 3:28 AM, Mel Gorman wrote: > This introduces a new sysctl vm.percpu_pagelist_high_fraction. It is > similar to the old vm.percpu_pagelist_fraction except it only adjusts > pcp->high to potentially reduce zone->lock contention while preserving > allocation latency when PCP lists have to be refilled. Look at me... Five patches later and I already forgot what the old one did and why it stinks. I wonder if you might do a wee bit of compare and contrast. Something like: The old vm.percpu_pagelist_fraction increased both the batch and high limits for the per-cpu page allocator. Its worst feature was that it led to absurdly large batch sizes that incurred nasty worst-case allocation latency. This new sysctl in comparison... Anyway, the approach looks sound to me. The batch size isn't important now, especially given the auto-scaling in patch 4. Acked-by: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
