<div dir="ltr"><div dir="ltr">On Thu, 9 Dec 2021 at 06:13, Julian Andres Klode <<a href="mailto:julian.klode@canonical.com">julian.klode@canonical.com</a>> wrote:<br></div><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi all,<br>
<br>
some time ago, the default compressor for initramfs was changed<br>
from lz4 -9 to zstd -19. This caused significant problems:<br></blockquote><div><br></div><div>Exactly three months later... we still haven't taken any action on this. Time to do something!</div><div><br></div><div>I have a few questions below but tl;dr: unless there are immediate objections, I'm going to make a change to initramfs-tools to allow the compression level to be configured and set the default to 12 for zstd.</div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
- it is very slow<br>
- it uses a lot of memory<br>
<br>
The former is a problem for everyone, the latter means that<br>
zstd just crashes on a Pi Zero.<br>
<br>
This is an analysis of what we have in terms of time spent,<br>
memory spent, and file size achieved, and where we can<br>
go from here.<br>
<br>
# Comparison of different compression levels<br>
<br>
## Desktop (ThinkPad T480s, jammy)<br>
<br>
level usertime elapsed memory fileSize<br>
lz4 9.65s 11.09s 12M 64M<br>
-1 5.69s 6.99s 24M 57M<br>
-6 12.59s 8.58s 99M 47M<br>
-12 19.85s 10.82s 249M 41M<br>
-19 71.29s 26.95s 519M 35M<br>
<br>
-> I believe that somewhere around -12 is a decent<br>
compromise between size and speed.<br></blockquote><div><br></div><div>I would agree that it's certainly better than -19.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
## Pi 4 (arm64, focal)<br>
<br>
Times have been measured for mkinitramfs only. A full<br>
update-initramfs call spends much more time copying<br>
some firmware bits to boot partition with flash-kernel<br>
<br>
level usertime elapsed memory fileSize<br>
lz4 21.10s 64.85s 21M 29M<br>
-1 13.73s 44.55s 21M 27M<br>
-6 26.07s 49.09s 91M 24M<br>
-12 48.18s 54.67s 203M 22M<br>
-19 130.07s 92.80s 350M 20M<br>
<br>
-> 6 is essentially free if the Pi 4 is idle. Nice.<br>
-> -6 is still 20% of total RAM of a Pi 0<br></blockquote><div><br></div><div>Are people really going to run an arm64 userland on a Pi 0?</div><div><br></div><div>Any "real" solution for pi 0 has to involve doing at least the bulk of the compression not on the pi, there's no real way around that. Which is something we should do, but realistically it's not happening for jammy.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
-> There's no meaningful difference between -6 and -12<br>
in terms of time elapsed. -6 uses 116% CPU, -12 uses<br>
145% CPU.<br>
<br>
## Adaptive compression<br>
<br>
zstd also supports adaptive compression, compressing as hard as<br>
it can while not impacting I/O speed. So hardware with slow I/O<br>
like a Pi would compress harder to avoid idling.<br>
<br>
This is somewhat suboptimal with recent update-initramfs though,<br>
as it first writes the cpio archive to the disk and then compresses<br>
it rather than doing it in a pipe where that would be more<br>
meaningful.<br>
<br>
Question: Does zstd --adapt adapt to memory available?<br></blockquote><div><br></div><div>While attractive, this does feel a bit risky. We want to be able to make reasonable predictions about the size of the initrd.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
# Way(s) forward<br>
<br>
To remedy the issue the proposal is to build with<br>
<br>
- zstd -1 on hardware with 512 MB or less memory<br>
- zstd between -1 and -19 on other hardware<br>
- zstd -19 during image building<br></blockquote><div><br></div><div>I think this broadly makes sense. I'll notice that currently initramfs-tools doesn't allow tuning the compression level at all :/ Probably fairly routine to add support for that though.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
Finding the right level between -1 and -19 is hard. The more<br>
cores you have, the less penalty you pay for higher level.<br></blockquote><div><br></div><div>You suggested -12 above. How about we try that to start with?</div><div><br></div><div>Do you have an idea how to detect "512 MB or less memory"?</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
Going for adaptive compression would remove the guess work, but<br>
will result in larger images on faster machines. Maybe that's<br>
fine, though - they probably have more space on /boot anyway?<br>
<br>
If we want to aim for 5% of total memory, we should probably<br>
aim for something like:<br>
<br>
-1 on <= 512MB<br>
-6 on <= 2 GB (or --adapt=min=1,max=6)<br>
-12 on the rest (or --adapt=min=12)<br>
<br>
It's clear that in all cases, zstd -1 is at least better than the<br>
lz4 -9 we used before; both in terms of space used, and time spent.<br></blockquote><div><br></div><div>Yeah that's interesting.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
# Concerns<br>
<br>
Lowering the compression level will reduce the boot speed by fractions<br>
of a second on hardware with fast I/O.<br></blockquote><div><br></div><div>Well speaking selfishly, the mean "number of boots per mkinitramfs call" for me is about 1...</div><div><br></div><div>Cheers,</div><div>mwh</div></div></div>