Messages in this thread | | | Date | Thu, 22 Nov 2018 15:31:03 +0900 | From | Minchan Kim <> | Subject | Re: [PATCH 4/6] zram: support idle page writeback |
| |
On Thu, Nov 22, 2018 at 03:15:42PM +0900, Minchan Kim wrote: > On Thu, Nov 22, 2018 at 02:40:40PM +0900, Sergey Senozhatsky wrote: > > On (11/22/18 14:04), Minchan Kim wrote: > > > > > > > additionally, it's too simple. It writes-back pages which can be > > > > swapped in immediately; which basically means that we do pointless > > > > PAGE_SIZE writes to a device which doesn't really like pointless > > > > writes. > > > > > > This patchset aims for *IDLE page* writeback and you can define > > > what is IDLE page by yourself. It doesn't do pointless writeback. > > > > > > > > It's a whole different story with idle, compressible pages writeback. > > > > > > I don't understand your point. > > > > Seems you misunderstood me. I'm not saying that IDLE writeback is bad. > > On the contrary, I think IDLE writeback is x100 better than writeback > > which we currently have. > > > > The "pointless writeback" comment was about the existing writeback, > > when we WB pages which we couldn't compress. We can have a relative > > huge percentage of incompressible pages, and not all of them will end > > up being IDLE: > > - we swap out page > > - can't compress it > > - writeback PAGE_SIZE > > - swap it in two seconds later > > I got what you mean now. Let's call it as "incompressible page wrieback" > to prevent confusing. > > "incompressible page writeback" would be orthgonal feature. The goal is > "let's save memory at the cost of *latency*". If the page is swapped-in > soon, it's unfortunate. However, the design expects once it's swapped out, > it means it's non-workingset so soonish swappined-in would be rather not > many, theoritically compared to other workingset. > If's it's too frequent, it means system were heavily overcommitted.
Havid said, I agree it's not a good idea to enable incompressible page writeback with idle page writeback. If you don't oppose, I want to add new knob to "enable incompressible page writeback" so by default, although we enable CONFIG_ZRAM_WRITEBACK, incompressible page writeback is off until we enable the knob. It would make some regressison if someone have used the feature but I guess we are not too late.
What do you think?
| |