On Tue, Jan 28, 2025 at 3:10 AM Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > On Mon, 27 Jan 2025 at 22:44, Joanne Koong <joannelkoong@xxxxxxxxx> wrote: > > > > Hi all, > > > > Recently, there was a long discussion upstream [1] on a patchset that > > removes temp pages when handling writeback in FUSE. Temp pages are the > > main bottleneck for write performance in FUSE and local benchmarks > > showed approximately a 20% and 45% improvement in throughput for 4K > > and 1M block size writes respectively when temp pages were removed. > > More information on how FUSE uses temp pages can be found here [2]. > > > > In the discussion, there were concerns from mm regarding the > > possibility of untrusted malicious or buggy fuse servers never > > completing writeback, which would impede migration for those pages. > > > > It would be great to continue this discussion at LSF/MM and align on a > > solution that removes FUSE temp pages altogether while satisfying mm’s > > expectations for page migration. These are the most promising options > > so far: > > This is more than just temp pages. The same issue exists for > ->readahead(). This needs to be approached from both directions. > I was assuming the cases for readahead and writethrough splice was going to be covered in the more generic mm session about which existing things in the system currently lead pages to be indeterminately unmigratable and which can be handled vs not. David, were you still planning to propose that as a topic? Maybe if there's a solution for the readahead/writethrough splice case from that, then that could also be applied to writeback too but if not, I think it might still be useful, given the non-trivial perf improvements we saw in the benchmarks, to align on whether there's any acceptable option that exists for removing writeback temp pages or whether we should abandon the attempt altogether. > This year I'll skip LSF but definitely interested in the discussion. > So I'll watch LWN for any updates :) This discussion about the FUSE writeback temp pages (and depending on that, the problem of how to integrate large folios with temp pages writeback state tracking) imo would seem more fitting as a BoF discussion as it seems pretty FUSE-specific. Maybe if there does end up being a discussion around this topic and it's a BoF, there could be an informal videocall option for it? Thanks, Joanne > > Thanks, > Miklos
