Re: Path forward for Virtualized Swap?

[Chris Li]

On Sun, Sep 20, 2026 at 2:36 PM Rik van Riel <riel@xxxxxxxxxxx> wrote:
>
> On Sun, 2026-09-20 at 14:18 -1000, Chris Li wrote:
> > On Sat, Sep 19, 2026 at 3:07 PM Rik van Riel <riel@xxxxxxxxxxx>
> > wrote:
> > >
> > > >
> > > > For our deployment, we only compress about 7-10% of the system
> > > > memory
> > > > before performance starts to suffer. The feedback loop will scale
> > > > it
> > > > back. I am very curious do you have actual work load to swap out
> > > > 100%
> > > > of the RAM.
> > >
> > > One common workload is a web browser with idle tabs.
> > >
> > > Those often compress pretty well, a ratio of 4:1 is
> > > not unheard of, which makes it easy to fill up
> > > 1/3 of memory with 133% of memory's worth of compressed
> > > memory.
> >
> > You did not answer my question. You are talking about user-space
> > processes (browsers). Do you run your browser in the data center, or
> > is that a laptop workload?
>
> The browser runs on the desktop, but the same
> kernel runs on both.

Interesting. The desktop is having a GUI or headless?

> The things we build here cannot be optimized for
> just one single workload. They have to work
> everywhere.

Sure, but on your desktop or laptop, what is the max amount of zswap
swapout compared to system RAM? It is obviously bounded somehow. What
is the actual number across all your deployments? You don't have to
give me an exact number; something with one digit of accuracy for the
maximum is good enough.

I am wondering why you keep dodging the question? It is just a number.
Do you not know the actual number, or do you not want to share it for
some unknown reason? I shared that our deployment commonly uses 7-10%
for zswap.

Chris