Ubuntu vs. Debian vs. openSUSE On The POWER9 Raptor Talos II

[phoronix]

Phoronix: Ubuntu vs. Debian vs. openSUSE On The POWER9 Raptor Talos II

While we frequently compare the performance of many x86_64 Linux distributions, we haven't done one under IBM POWER9 since getting our hands on the Raptor Computing Systems' Talos II back in November. It's been very interesting to benchmark this libre hardware that's high performance with having 44 cores / 176 threads at 3.80GHz. But how much more performance can be tapped by using other Linux distributions? Here's a look with some of the current POWER9 Linux distribution options.

http://www.phoronix.com/vr.php?view=27486

[Compartmentalisation]

Can you please write some things about the results? Yes I see which is faster when and where, but why is Debian 9 sometimes faster than Debian 10 for instance?

[Michael]

Can you please write some things about the results? Yes I see which is faster when and where, but why is Debian 9 sometimes faster than Debian 10 for instance?

When Debian 9 comes out faster than Debian 10 is usually when the older versions of e.g. GCC or Golang are faster than newer versions, such as for compile times being faster on older releases due to less optimization passes. In other cases there are performance regressions or other issues at play.

[sharkcz]

I would be interested in knowing more about the issues you had when installing Fedora. What version/media/method did you use?

[Michael]

I would be interested in knowing more about the issues you had when installing Fedora. What version/media/method did you use?

Basically it was unresponsive as soon as hitting the installer. Tried Fedora Server 29 spin. At first was using ISO to USB drive and then booting from DVD.

[sharkcz]

Basically it was unresponsive as soon as hitting the installer. Tried Fedora Server 29 spin. At first was using ISO to USB drive and then booting from DVD.

Both ways should work in Fedora 30, there was a long time issue with installs from USB, that has been fixed. I've also tested install from physical DVD and it was OK too. Don't hesitate contact me even directly if you have questions about Fedora for Power, it's part of my job.

[andyprough]

When running a default installation of openSUSE, you should check to see if baloo is running, which it normally is out of the box. If so, you will get strange hang-ups due to baloo almost immediately starting to index the entire file system. As I've commented before, a default openSUSE Tumbleweed setup is an odd thing to benchmark, since almost no one in reality would run it that way for the purposes for which you benchmark it. Not to mention the btrfs vs ext4 decision.

Other than that, lot of great looking data here.

[Michael]

When running a default installation of openSUSE, you should check to see if baloo is running, which it normally is out of the box. If so, you will get strange hang-ups due to baloo almost immediately starting to index the entire file system. As I've commented before, a default openSUSE Tumbleweed setup is an odd thing to benchmark, since almost no one in reality would run it that way for the purposes for which you benchmark it. Not to mention the btrfs vs ext4 decision.

Other than that, lot of great looking data here.

Does openSUSE install Baloo even on server installs?

[andyprough]

Does openSUSE install Baloo even on server installs?

I have no idea what you are running Michael. Are you running a headless server or what? openSUSE installs baloo with KDE and it runs out of the box by default, and it will severely mess up your results. So if you have a desktop, you probably have baloo running and indexing files for the first day or so of installation.

As I said, normal users don't run it with default settings, and they certainly aren't useful for benchmarking purposes. First thing to do would be turn off baloo, and then if you want indexing, maybe let it run overnight while you are sleeping.

[Michael]

I have no idea what you are running Michael. Are you running a headless server or what? openSUSE installs baloo with KDE and it runs out of the box by default, and it will severely mess up your results. So if you have a desktop, you probably have baloo running and indexing files for the first day or so of installation.

As I said, normal users don't run it with default settings, and they certainly aren't useful for benchmarking purposes. First thing to do would be turn off baloo, and then if you want indexing, maybe let it run overnight while you are sleeping.

This was a POWER9 openSUSE server install without any desktop, so was just wondering if openSUSE still installs baloo or other KDE components in this case for any strange reason.