On 3/4/24 13:35, Yin, Fengwei wrote: > Even the file read throughput is almost same. But the distribution of > order looks like a regression with ab4443fe3ca6 (more smaller order > page cache is generated than parent commit). Thanks. There may be confusion here. Let me clarify it as: I shouldn't say folio order distribution is a regression. It's smaller folio order cause more folios added in page cache for same workload. And raise the lru lock contention which trigger the regression. Regards Yin, Fengwei
