Support For Compressing The Linux Kernel With LZ4

[phoronix]

Phoronix: Support For Compressing The Linux Kernel With LZ4

A set of patches that allow the Linux kernel image to be compressed with the LZ4 lossless compression algorithm have been published. The size of LZ4-compressed Linux kernel images are larger than using LZO compression, but there's promise that the boot times could be better...

http://www.phoronix.com/vr.php?view=MTI4NjM

[BO$$]

What is the point of having a compressed kernel? Faster loading times from the HDD?

[Ericg]

What is the point of having a compressed kernel? Faster loading times from the HDD?

Smaller footprint which means: yes faster loading times from hdd, smaller filesize ON the hdd (embedded) and means it can be shoved into RAM if you wanted the entire live system into ram for responsiveness

[mercutio]

what happened to btrfs's lz4 compression support?

it seems lz4 makes sense a lot for always compressing files when not using a sandforce ssd.

[AJenbo]

I wonder what would happen if a lossy algorythem was used

[bnolsen]

I wonder what would happen if a lossy algorythem was used

then you have a lossy system.

[Ericg]

I wonder what would happen if a lossy algorythem was used

Then i'd hope that the decompression algorithm was very good about reconstructing the lost data which would probably mean a longer decompression time =P

[mark_]

come on... 100ms difference, you cannot even measure that with your watch. But you can measure the init-time after the kernel is loaded...

[oliver]

Smaller footprint which means: yes faster loading times from hdd, smaller filesize ON the hdd (embedded) and means it can be shoved into RAM if you wanted the entire live system into ram for responsiveness

I don't think you can have a running kernel, compressed in ram. As far as I know, it gets loaded from some form of storage (flash, hdd, nfs) and decompressed into ram.

come on... 100ms difference, you cannot even measure that with your watch. But you can measure the init-time after the kernel is loaded...

This was on their test-system I'm sure. I bet on an ARM-m3 it takes quite a lot longer. So we're talking about 100% faster decompression times (150ms vs 300ms).

[newwen]

I guess Google is also very interested on this getting into mainline kernel.