On Fri, Sep 22, 2023 at 05:57:40PM +0000, Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote: > memcg_rstat_updated() uses the value of the state update to keep track > of the magnitude of pending updates, so that we only do a stats flush > when it's worth the work. Most values passed into memcg_rstat_updated() > are in pages, however, a few of them are actually in bytes or KBs. > > To put this into perspective, a 512 byte slab allocation today would > look the same as allocating 512 pages. This may result in premature > flushes, which means unnecessary work and latency. > > Normalize all the state values passed into memcg_rstat_updated() to > pages. I've dreamed about such normalization since error estimates were introduced :-) (As touched in the previous patch) it makes me wonder whether it makes sense to add up state and event counters (apples and oranges). Shouldn't with this approach events: a) have a separate counter, b) wight with zero and rely on time-based flushing only? Thanks, Michal
Attachment:
signature.asc
Description: PGP signature
