On Wed, Jan 20, 2021 at 10:59:52AM -0800, Samuel Sieb wrote:
> That's about changing settings. When restarting the zram service,
> it has to remove the swap device which forces all the swapped pages
> back into RAM. This might not be successful if you don't have
> enough RAM, so if you are using
On 1/20/21 8:14 AM, Matthew Miller wrote:
On Sun, Jan 17, 2021 at 05:58:45PM +, Zbigniew Jędrzejewski-Szmek wrote:
Someone on Fedora Discussion¹ noticed this number and wonders if it's meant
to be 8192. Not that it probably matters very much, but, you know, nice
round numbers and all.
Yep,
On Sun, Jan 17, 2021 at 05:58:45PM +, Zbigniew Jędrzejewski-Szmek wrote:
> > Someone on Fedora Discussion¹ noticed this number and wonders if it's meant
> > to be 8192. Not that it probably matters very much, but, you know, nice
> > round numbers and all.
>
> Yep, it was. I'll fix the wiki.
O
On Sun, Jan 17, 2021 at 12:28:10PM -0500, Matthew Miller wrote:
> On Mon, Jan 11, 2021 at 11:50:03AM -0500, Ben Cotton wrote:
> > == Scope ==
> > * Proposal owners: change the defaults in `zram-generator-defaults`
> > package to `zram-fraction=1.0`, `max-zram-size=8096`.
>
> Someone on Fedora Disc
On Mon, Jan 11, 2021 at 11:50:03AM -0500, Ben Cotton wrote:
> == Scope ==
> * Proposal owners: change the defaults in `zram-generator-defaults`
> package to `zram-fraction=1.0`, `max-zram-size=8096`.
Someone on Fedora Discussion¹ noticed this number and wonders if it's meant
to be 8192. Not that i
On Tue, Jan 12, 2021 at 04:58:36PM +0100, Vitaly Zaitsev via devel wrote:
> On 11.01.2021 21:55, ElXreno wrote:
> >Also, other data (video editors, for example) which are not
> >compressible can get into RAM.
>
> Windows guests running under KVM/qemu too.
>
> Windows 10 uses its own build-in memo
It seems I was wrong. I did the same test again, and this time even with
more `zram-fraction'. The Darktable memory is compressed, there was no
more hang-up problem. Sorry about the noise, maybe it was a regression
in the old kernel, or my configuration error somewhere.
On 1/12/21 5:55 PM, Mat
Some tests: https://youtu.be/d4Sc80TMEtA
webkit2gtk3 compilation with zram-fraction=1, max-zram-size=8192
Physical RAM: 8GB.
Kernel 5.10.6
Fedora 33, with such deviations from default:
- zram-fraction=1, max-zram-size=8192 (this is default for f34)
- le9 patch - https://github.com/hakavlad/le9-pa
> system hangs with earlyoom and swap on zram, disksize=150% MemTotal,
compression ratio at the end is 1.4, SwapFree=25%, the zram device occupies
80% of the memory with running Blender https://imgur.com/a/wNvBVIn.
> With disksize=100% MemTotal this should be very rare case.
-- https://pagure.io/
On 11.01.2021 21:55, ElXreno wrote:
Also, other data (video editors, for example) which are not compressible
can get into RAM.
Windows guests running under KVM/qemu too.
Windows 10 uses its own build-in memory compression and deduplication
service.
--
Sincerely,
Vitaly Zaitsev (vit...@eas
On Tue, Jan 12, 2021 at 4:06 AM Chris Murphy wrote:
>
> On Mon, Jan 11, 2021 at 1:55 PM ElXreno wrote:
>
> > I think it's a bad idea. It is better to make `zram-fraction` equal to
> > 0.8 or 0.9, but not 1.0.
>
> Just to clarify. zram-fraction 1.0 means the zram *device size* will
> be equal to R
On Mon, Jan 11, 2021 at 11:55:12PM +0300, ElXreno wrote:
> For example, if you run Darktable, load a floating-point TIFF image
> into it, and try to export it, on systems with little RAM, it will
> fill both RAM itself and zram, and cause the system to freeze
> completely because the memory is almo
Different question: This sounds like you're handling data that won't
fit into RAM+Swap anyway, using zram isn't going to help (or hurt),
because with or without compression, your data just doesn't fit in
memory.
I'm talking about possible outcomes, including the ones I've had.
What does the maxi
On Mon, Jan 11, 2021 at 9:55 PM ElXreno wrote:
>
> I apologize for the possible duplicate message, the past didn't show up
> for reasons I don't understand.
>
> I think it's a bad idea. It is better to make `zram-fraction` equal to
> 0.8 or 0.9, but not 1.0.
> For example, if you run Darktable, lo
Is earlyoom disabled on this system? It should send SIGTERM to
something in this case - which is not the best UX as a hint that an
adjustment needs to be made. But I want to make sure there isn't
something else going wrong.
Yes, it's off. And I have an assumption that earlyoom may not have time
t
On Tue, Jan 12, 2021 at 12:14 AM ElXreno wrote:
>
> > Does this reduce your concern?
> In a way, yes. Most of the time memory will be compressible, I agree,
> but there are non-compressible data, and I'm only concerned about that.
>
> In my opinion, it's better not to set the zram size to 100% for
Does this reduce your concern?
In a way, yes. Most of the time memory will be compressible, I agree,
but there are non-compressible data, and I'm only concerned about that.
In my opinion, it's better not to set the zram size to 100% for
everyone. Whoever is sure that their data will be compres
On Mon, Jan 11, 2021 at 1:55 PM ElXreno wrote:
> I think it's a bad idea. It is better to make `zram-fraction` equal to
> 0.8 or 0.9, but not 1.0.
Just to clarify. zram-fraction 1.0 means the zram *device size* will
be equal to RAM, capped at 8G. The actual amount of memory used is
zero, until t
I apologize for the possible duplicate message, the past didn't show up
for reasons I don't understand.
I think it's a bad idea. It is better to make `zram-fraction` equal to
0.8 or 0.9, but not 1.0.
For example, if you run Darktable, load a floating-point TIFF image into
it, and try to export
https://fedoraproject.org/wiki/Changes/Scale_ZRAM_to_full_memory_size
== Summary ==
Fedora 33 [[Changes/SwapOnZRAM|enabled zram by default]]. The size of
the virtual swap devices was set so that the amount of memory used for
compressed swap pages was limited to a quarter of physical memory in
typ
20 matches
Mail list logo
