Re: [PATCH v5] zswap: memcontrol: implement zswap writeback disabling

[Nhat Pham]

On Tue, Nov 21, 2023 at 11:09 AM Chris Li <chrisl@xxxxxxxxxx> wrote:
>
> On Tue, Nov 21, 2023 at 10:13 AM Nhat Pham <nphamcs@xxxxxxxxx> wrote:
> >
> > Hi Chris!
> >
> > Thanks for the patch. Would you mind if I spend some time staring
> > at the suggestion again and testing it some more?
>
> Of course, by all means. That is just the minimal version to be
> functional compatible with your zswap.writeback.
>
> I might consider a follow up patch to add "no_zswap" and "none" to
> convert the SSD only swapfile, which can't be expressed by
> zswap.writeback.
> That should cover all 4 combinations of zswap and swap files without
> creating a custom swap tiers list.
>
> "all": zswap + swapfile
> "zswap": zswap only
> "no_zswap": swapfile only.
> "none": no swap.
>
> All keyword names are open to suggestions.

SGTM! There might be some functionality duplication between
memory.swap.tiers = no_zswap and memory.zswap.max = 0, but
otherwise this seems reasonable to me.

no_zswap sounds a bit awkward, but I can't come up with a better
name.

>
> >
> > If everything is good, I'll squash this patch with the original version,
> > (keeping you as a co-developer of the final patch of course), and
> > update the documentation before re-sending everything as v6.
>
> Great!
>
> >
> > Anyway, have a nice Thanksgiving break everyone! Thanks for
> > taking the time to review my patch and discuss the API with me!
>
> My pleasure to discuss the swap with you. We should do the online
> "swap meet" and invite other developers who are interested in the swap
> area as well.

I look forward to this meeting! I'd love to discuss more about (z)swap
development (and more generally, multi-tier memory management).

Generic page promoter/demoter that takes into account workload
(cgroup), access recency (LRU + generations)/frequency, and tier
characteristics (latency, bandwidth, etc.) will be awesome to explore!

>
> Chris