Firefox memory usage

[Ralf Mardorf]

On Mon, 9 Nov 2020 15:19:15 +0100, Uwe Laverenz wrote:
>Maybe the slowdowns you see on Intel CPUs are due to the many fixes
>that became necessary because of the spectre an meltdown side-channel 
>problems. AFAIK Intel CPUs are affected more than AMD CPUs.

I don't think so, more likely X11 and GTK related issues might be
culprits. OTOH a lot of issues didn't happen before Meltdown and
Spectre _and_ just disabling all mitigations not necessarily brings
back old fast paths.

Due to the mitigations Linux lost probably one or the other fast path,
but it seems not to have that much impact on resources or performance.
The mitigations seem to be mostly provided by the kernel, not by the
microcode, so it indeed could be different on FreeBSD. However, some
Intel and X users experience issues with new Linux kernels, that are
seemingly related to the graphics driver. If I'm using a Linux kernel
5+ and launch Firefox, then the graphics does freeze for several
seconds and the machine is unresponsive for this seconds. Other
experience crashes that force them to reboot. This issue does effect
other software, too, but Firefox is most prone to cause this issue on
my machione. I stay with Linux 4+ to get rid of this issue. There's
still another X related issue common on some Linux machines. Apps tend
to crash all the times with `!xcb_xlib_threads_sequence_lost'. Mostly
GTK2 apps are affected and often migrating to a GTK3 branch of the same
app solves the issue, but sometimes it happens to GTK3 apps, too.

Too funny, I planed to completely migrate to FreeBSD to get rid of those
new Linux issues, but seemingly it's all the same when using FreeBSD.
OTOH I'm not surprised, since when opening a GTK3 app even on the
fastest, modern computers, the window shows the GTK widgets first as
visible black boxes, it's at least noticeable as a kind of flash.

Btw. I migrated from AMD to Linux a while back, since all my machines
were AMD and a PITA, at least for pro-audio usage. With my Intel machine
I got rid of most, if not all issues, but a while ago upstream step by
step does introduce new issues.

To some extend it's better when using proprietary software on iPadOS.
By paying upstream does remove one bug by another and serious bugs
usually don't last for long. Upstream of gratis FLOSS software often
tends to put the blame on other and then to close a bug report.