Zstd'ing The Kernel Might See Mainline With Linux 5.9 For Faster Boot Times

[phoronix]

Phoronix: Zstd'ing The Kernel Might See Mainline With Linux 5.9 For Faster Boot Times

It looks like the long ongoing work for compressing the Linux kernel image with Zstd might finally soon be mainlined, potentially for next month's Linux 5.9 cycle kicking off as the "v6" patches sent out this week were done as a Git pull request...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-Zstd-Compression-Close

[Danny3]

Didn't Ubuntu already have done something like this ?
I don't know about others, but I would prefer faster boot speed than saved space.
There are many case where I need to boot the laptop or desktop and find some information as fast as possible like looking for an address on a map or something from the computer itself.

[yoshi314]

Didn't Ubuntu already have done something like this ?
I don't know about others, but I would prefer faster boot speed than saved space.

smaller (compressed) kernel with fast unpack speed = faster boot. zstd is designed to decompress quickly, pretty much on par with gzip. it actually seems to pull ahead in decompression in benchmarks.


on low-end embedded hardware things might be different, though.

[rene]

yep, that's why #t2sde switched to zstd for everything, packages, kernel & initrd since three years already ;-) https://www.youtube.com/watch?v=dQee4IDoUow

[skeevy420]

About time.

[microcode]

Zstd can decompress faster than lz4 at some equivalent ratios for some data. It is really impressive!

[discordian]

smaller (compressed) kernel with fast unpack speed = faster boot. zstd is designed to decompress quickly, pretty much on par with gzip. it actually seems to pull ahead in decompression in benchmarks.


on low-end embedded hardware things might be different, though.

The difference might be smaller, but I'd guess that it will beat everything but lz4/lzo in speed even on microcontrollers.
Its design omits branches whenever possible, that's a win everywhere, the only real downside to alternatives would be the comparable huge decompression code (I did not manage to get this below 64K) - which is the knockout argument for microcontrollers.

[evergreen]

(I did not manage to get this below 64K)

I managed to get the decompression code down to ~15 KB with just a few macro definitions and compilation flags (documented on the project's page).

[boeroboy]

Not sure if pzstd parallel decompressionthreads are an option before the kernel is running but wow looking forward to testing 5.9.

[oibaf]

Would be nice to update the zstd lib in the kernel, still pretty old and missing latest improvements I think.