Intel Linux Kernel Optimizations Show Huge Benefit For High Core Count Servers

[Linuxxx]

While the throughput performance of Intel's Clear Linux is certainly impressive, note that it is using both the performance governor & THP (Transparent HugePages) by default, unlike Ubuntu.

BTW, RHEL 9 has also copied these settings from Clear Linux, so don't be surprised when it leads over Ubuntu in the next round of benchmarks Michael is going to publish.

Just stating it here in advance before the "Ubuntu is the reason why my life is ruined" hatewagon starts rolling in...

[Mark Rose]

I'm pretty sure that's 960 thread, but 480 core.

Yes, threads, my bad.

[Michael]

Would like to see DragonFlyBSD thrown in since it claims to be better than other kernels in high SMT situations. See if it dips as much as Linux does at high core counts. I must say this was shocking as a discovery. Never dreamed 60 cores would preform better than 120 cores or 240 threads!

Last time I tried in ~January, DragonFlyBSD stable release couldn't boot on the Eagle Stream server.

[kylew77]

Last time I tried in ~January, DragonFlyBSD stable release couldn't boot on the Eagle Stream server.

Darn the luck Michael, thanks so much for trying though!

[npwx]

Its a bit unfortunate that the comparison was done between Ubuntu and Clear Linux, instead of just a vanilla Kernel vs a Clear linux kernel. This throws a million unknown factors into the mix.

[F.Ultra]

Would like to see DragonFlyBSD thrown in since it claims to be better than other kernels in high SMT situations. See if it dips as much as Linux does at high core counts. I must say this was shocking as a discovery. Never dreamed 60 cores would preform better than 120 cores or 240 threads!

It's a quote normal and quite expected side effect of the CPU having 60 physical cores and the applications tested are using a shared memory model. This have nothing to do with the OS and everything to do with arjan_intel wrote in https://www.phoronix.com/forums/foru...48#post1380648

[waxhead]

What continues to surprise me with benchmarks such as these is that Ubuntu is "always" the default go-to distro. WHY????! While Ubuntu may be popular I would argue that it is not necessarily a sane choice. (And Ubuntu != Debian just for the record)

The absolute largest mother distros out there is essentially Debian and RedHat which forms a base for almost every other distro out there, so why not test on those first and foremost?!

And my all means, there is nothing wrong in throwing other distros into the mix either and compare how they potentially are better/worse than the mother distro from which they came, but I think benchmarks would make a lot more sense, and would be easier to read-/compare with- ...if they did use a common base that was always the same. And in this case it would be really interesting to see how Clear Linux compares against both pure RHEL and Debian (stable and testing)

(and yes, I am a pure Debian-ist myself so I am a bit biased and admit to being rather anti-Ubuntic, but I still think my point is valid despite this!)

[coder]

What continues to surprise me with benchmarks such as these is that Ubuntu is "always" the default go-to distro. WHY????! While Ubuntu may be popular I would argue that it is not necessarily a sane choice. (And Ubuntu != Debian just for the record)

If he compares everything to Ubuntu, then you can more easily infer relative performance between any other pair of distros that he's benchmarked against it. It makes sense for him to settle on a common baseline, and Ubuntu doesn't seem a bad choice -- especially now that Centos is defunct.

The absolute largest mother distros out there is essentially Debian and RedHat which forms a base for almost every other distro out there, so why not test on those first and foremost?!

As for using Debian, doesn't Ubuntu tend to have more patches and hardware enablement? It certainly has more official vendor support than Debian, which is key for many of his tests.

Regarding Redhat -- which Redhat do you mean? RHEL? Most of us are too cheap to spring for a license. AFAIK, Fedora is a rolling distro, and therefore not a good choice of a common baseline. Centos is gone, and none of its spiritual successors seems to have the same traction.

[AlanTuring69]

What continues to surprise me with benchmarks such as these is that Ubuntu is "always" the default go-to distro. WHY????! While Ubuntu may be popular I would argue that it is not necessarily a sane choice. (And Ubuntu != Debian just for the record)

The absolute largest mother distros out there is essentially Debian and RedHat which forms a base for almost every other distro out there, so why not test on those first and foremost?!

And my all means, there is nothing wrong in throwing other distros into the mix either and compare how they potentially are better/worse than the mother distro from which they came, but I think benchmarks would make a lot more sense, and would be easier to read-/compare with- ...if they did use a common base that was always the same. And in this case it would be really interesting to see how Clear Linux compares against both pure RHEL and Debian (stable and testing)

(and yes, I am a pure Debian-ist myself so I am a bit biased and admit to being rather anti-Ubuntic, but I still think my point is valid despite this!)

Ubuntu is far more common in the enterprise, cloud etc because it stays up to date with security and performance fixes while being extremely stable. I encounter Ubuntu regularly while only encountering Debian a handful of times. Arguments can be made about Ubuntu's desktop choices but it is genuinely a top performer in the server world and does not have the same questionable decisions. The reality is this *-ist opinion does not matter for people trying to do their job with thousands of server instances and they are going to pick the tool that works best regardless of these "Debian-ist" "anti-Ubuntic" blah blah bullshits that are largely based on ideals instead of practicality.

[arjan_intel]

Its a bit unfortunate that the comparison was done between Ubuntu and Clear Linux, instead of just a vanilla Kernel vs a Clear linux kernel. This throws a million unknown factors into the mix.

question would be: Which configuration for the vanilla kernel?
Some of the scaling things are very kernel config dependent, and for that, using Ubuntu is quite reasonable to compare