Some Of The Workloads Still Seeing Lower Performance On Linux 5.5 Git

[phoronix]

Phoronix: Some Of The Workloads Still Seeing Lower Performance On Linux 5.5 Git

Last night I shared the results from what's causing one of the performance regressions in Linux 5.5 but sadly more regressions remain that are currently being tracked down...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-5.5-Regressions-Past-AA

[pyler]

Yet I havent seen any performance improvement due to new scheduler code. Linus should revert it really until it *works*.

[wooque]

You should be part of kernel development, I can't comprehend they don't have some kind of continuous performance testing as part of development to catch this kind of issues.

[milkylainen]

I commend your endless hard work. But it seems I revisit this topic like a broken record.
It's sad that Linux has no real (validated) performance regression testing.

This is just another one...

[timofonic]

Is this a joke? How pathetic, no performance regression testing and improvements at all.

[PuckPoltergeist]

Yet I havent seen any performance improvement due to new scheduler code. Linus should revert it really until it *works*.

Why? There is still no evidence, that the scheduler-rework is causing the regressions, Michael sees.

[birdie]

Is this a joke? How pathetic, no performance regression testing and improvements at all.

It's not a joke, it's open source. You must be new to it if you're surprised one tiny bit.

Regressions is basically an Open Source synonym.

Proper regression testing takes a lot of resources, time and money and even Microsoft has kinda worsened in this regard after they released Windows 10 which is basically perpetual beta software (aside from Windows 10 LTSC).

[birdie]

Michael

You must absolutely write to LKML (and CC the responsible developers) and file a bug report at https://bugzilla.kernel.org/ 'cause otherwise your hard word will be all for naught.

Speaking of LKML: if you don't get any replies within 48 hours after sending a message, you should send it again.

[PuckPoltergeist]

Regressions is basically an Open Source synonym.

That's bullshit. You have the same problems with closed source. The kind of development model doesn't impact software bugs here.

[FireBurn]

If the PTS is capable of bisecting these issues using graphs and what not, why not get it to submit the bug report automatically directly to bugzilla

I know the Gentoo tinderbox used to submit bug reports to the Gentoo bugzilla so I know this automation is possible