On Mon, Apr 24, 2023 at 8:41 AM Logan Gunthorpe <logang@xxxxxxxxxxxx> wrote: > > > > On 2023-04-24 00:45, Christoph Hellwig wrote: > > On Thu, Apr 20, 2023 at 02:10:02PM -0600, Logan Gunthorpe wrote: > >>> I am hoping to make raid5_make_request() a little faster for non-rotational > >>> devices. We may not easily observe a difference in performance, but things > >>> add up. Does this make sense? > >> > >> I guess. But without a performance test that shows that it makes an > >> improvement, I'm hesitant about that. It could also be that it helps a > >> tiny bit for non-rotational disks, but we just don't know. > >> > >> Unfortunately, I don't have the time right now to do these performance > >> tests. > > > > FYI, SSDs in general do prefer sequential write streams. For most you > > won't see a different in write performance itself, but it will help with > > reducing GC overhead later on. Yeah, this makes sense. > > Thanks. Yes, my colleague was able to run performance testing on this > patch and didn't find any degradation with Jan's optimization turned on. > So I don't think it's worth doing this only for rotational disks and > Jan's original patch makes sense. I will ship Jan's original patch. Thanks, Song
