Linux 3.5 Through Linux 4.4 Kernel Benchmarks: A 19-Way Kernel Showdown Shows Some Regressions

[haagch]

Agreed. Still waiting for the Liquorix build of 4.4 though.

Linux-zen 4.4 has been ready for a while. Suspending my laptop with linux-zen 4.4 from archlinux didn't work for me though, so I'm using mainline 4.4 now again. The difference is hardly, if at all, noticeable anyway.

[F1esDgSdUTYpm0iy]

Suspending my laptop with linux-zen 4.4 from archlinux didn't work for me though

Interesting phenomenon; I'd have expected latency related tuning not to affect things like suspension.

[Shimon]

When is the search functionality coming back? I'd like to see a few kernel 2.4 results

I've got a partition like that on my old system but couldn't compile php with zlib support or I'd have posted my results a while ago.

[Kano]

I only tested 4.2 and 4.4 in a modified Ubuntu config (AHCI=y) without initrd and 4.4 needed at least 0.4-0.5 s longer for systemd-analyze. There might be problems with parport for some systems too with 4.4.

[tigerroast]

The benchmarks definitely help paint the big picture. Glad to see FLAC encoding didn't regress though.

It's probably an exercise in futility to have someone directly work on the kernel specifically to increase performance simply because performance seems to be the effect as opposed to the cause. More bug-squashing would be nice, but I'm from the outside looking in. Therefore, I can't really speak on that matter.

Squashing kernel bugs must be a nightmare.

[F1esDgSdUTYpm0iy]

Squashing kernel bugs must be a nightmare.

And, it is! Well, at times anyhow.

[DeFTeR]

That were eye-opening results. Thanks for that benchmarks.Seems like it is time to downgrade to 3.13 for now.

[F1esDgSdUTYpm0iy]

Or upgrade to 4.4. Since it is essentially on-par with 3.13's performance, even exceeding it in a few tests.

[slacka]

Any theories as to what change might have caused the Compile Bench's performance regressions between 3.8 and 3.10? Linux 4.4 is still considerably slower than 3.8. I didn't see a bug report at https://bugzilla.kernel.org. Should we file one?

[F1esDgSdUTYpm0iy]

Should we file one?

I think bisection should be the first step; to determine the exact cause of the issue. However, this could be quite a lengthy process.