Ubuntu 18.04 vs. 20.04 LTS Performance Preview | Hacker News Reader

Ubuntu 18.04 vs. 20.04 LTS Performance Preview

phoronix.com

142 pointsarnaudsm49 comments

Open articleView on HN

No explanation why Ubuntu is 10-30% faster in most of the benchmarks? That's a lot.

Maybe something to do with better spectre/meltdown mitigations in the newer kernel?

Page one of the article explains how the rig was configured, noting various security things.

https://www.phoronix.com/scan.php?page=article&item=xeonr-ub...

https://openbenchmarking.org/embed.php?i=2003242-NI-XEONCASC...

It's worth keeping in mind that they are running benchmarks on a massive dual-socket 48 core machine where slight tweaks of the scheduler can have a big effect. I would be very surprised to see such big differences on a normal desktop or server.

What's a little confusing is that they are not testing the same versions of the software but (presumably) the distro default version. So these benchmarks are not useful to test kernel and lower level performance differences.

Their goal was testing two versions of Ubuntu (as per the title), that includes every difference between them. Example: Ubuntu 18.04 upgrades Linux up to version 5.3, Ubuntu 20.04 comes with 5.4. There are several other upgrades. They are listed in the first page of the post.

Most of it seems to be down to newer packages and newer compiler versions.

Those are part of the release, otherwise you'd better test different kernels on 18.10.

Would this possibly explain some of the speed improvements?

https://arstechnica.com/gadgets/2019/11/ubuntu-19-10-quite-s...

That's awesome. Also very cool to see u2f/fido support for ssh.

https://wiki.ubuntu.com/FocalFossa/ReleaseNotes

Awesome! Love my yubikey.

YEAH!! Finally. Trying to assemble my own rube goldberg machine for key access is exhausting.

What? What do you mean? Could you show an explanation of what you're trying to achieve? Genuinely interested.

Using a hardware token as the authenticator for SSH sessions. Taken to the fullest, this allows you to completely avoid password authentication for SSH, except for the initial upload of your public key.

Thanks! That makes sense. I don't get the name, but this is what we call that now?

U2F/Fido refers mainly to doing through the browser, but as you probably know, the browser is only the end of the line of a long chain of usability oriented changes. It's been a decade in the making. Now SSH is using that mechanism, just not through the browser. Personally I'm psyched, lots of apps can start using this approach and get a great user experience!

It looks like it's not terribly portable (I can't walk up to a machine, plug my key in and SSH home), but it's really easy to set up with one command.

Does it only support ecdsa? Surprised they don't use ed25519 in their example.

Edit: nope, seems ed25519 supported too https://manpages.ubuntu.com/manpages/focal/en/man1/ssh-keyge... although internet seems to suggest hardware support is far more limited. Tried generating one on MacOS and indeed my Yubikey seems unsupported.

Unfortunately, you can't use Ed25519 keys unless you have the latest/newest model(s) of Yubikey and that Yubikey came with at least a specific firmware version (5.2.3, IIRC).

Has anyone benchmarked this against https://clearlinux.org/ ?

EDIT 1: To answer my own question: https://www.phoronix.com/scan.php?page=article&item=ubuntu20...

With a pre release Ubuntu 20.04 Clear Linux is slightly faster.

EDIT 2: tldr; AMD 8% faster in Clear Linux, Intel 14% faster. AMD is miles faster (20%) than Intel in almost everything on Linux.

ClearLinux is the dark horse doing amazing things that the HN community seems to shun.

Where other distros then learn from so people not willing to go with Intel's toy project can enjoy them.

I have no idea if it is a toy project or not? Do you have evidence things are often broken for example?

Toy in the sense that it's fully at Intel's whims.

So Ubuntu is also a Toy project because it's fully at Ubuntu's whims?

Well, they are not a chip giant, so they actually have make money off of it. Intel could drop this project anytime, and not see any difference in revenue, and many have questioned the backstory of this distro.

I, for one, plead guilty of shunning here.

The homepage states that the desktop flavor is aimed at developers ('Clear Linux OS offers options that are tailored specifically for the developers’ production'), then searching for well known tiling WM (xmonad, i3) on the packages page yields not result. Not sure what other big things are left apart too. No thanks.

They apparently have sway if you are willing to jump ship to wayland.

The geometric mean across the tests is 25% faster. That shouldn’t happen. Was there something wrong with 18.04?

gnome

Some of the specific performance enhancements are quite remarkable. Almost double the performance with PHP? Wow, that's quite the optimization for a long-lived (legacy?) technology. And the improvements in ML number crunching are in some cases just as impressive. Remarkable work within the community to wring more performance.

How feasible would it be to run a current Ubuntu on an older i5 with 4GB of RAM?

Should run fine, especially if you use an SSD. You're free to use any of the lightweight desktop environments (lxde, xfce, mate, etc.). The remainder depends on your workload/usecase wrt. to mainly RAM-requirements for tabbed browsing (or electron apps).

The RaspberryPi 4 also only has up to 4GB of RAM and is capable of driving 2 4K screens.

That IS my daily driver. A quad-core i5, but I can't imagine a dual core struggling that much. 4GB of ram is fine for any linux distro itself but might not be good enough for a browser-centeric workflow. I get around that by using extensions like Tab-Discard or TheGreatSuspender to "pause" unused tabs.

Using a userland oom-killer does help a lot with deadlock situations (https://news.ycombinator.com/item?id=22833471)

I wouldn't still recommend Gnome (Default Ubuntu), any other desktop would be lighter on RAM and CPU. KDE, MATE, or Budgie are all great.

I do have a speedy SSD though, when I was on my HDD the pill was quite harder to swallow.

This has been my experience as well. I do a lot of work on a computer running 18.04 with an i5-5300U (2 core, 4 thread), and the change between having an SSD instead of an HDD was probably the single largest performance improvement I've had because of a component change.

I just tried out both Ubuntu (gnome) and Mint (Cinnamon) on a 2015 MacBook Air, dual core i5 with 4 GB RAM. Mint was noticeably snappier - I really didn't enjoy Ubuntu, but a lighter desktop will probably make the difference.

Another comment talked about the RPi 4's capabilities, but mine had multi-second input lag running a fresh install of Ubuntu/Gnome on a single 4k screen.

RPi4 had significant, hardware-based performance issues in v1 of the board...

Do you have any links to this? I know about the USB-PD issue which they fixed in v1.2, but that's about it.

I think it was all bound to this - power issues make weird quirks throughout the system.

The difference is astounding in virtualbox. Ubuntu Gnome is nearly unusable- mint works pretty well. Using i3 for Ubuntu helps quite a bit.

With i3 (no GNOME elements at all, and precious little desktop environment stuff), I think my old 1920×1080 laptop tended to idle at using around 60MB of RAM. With 8GB of RAM and a 128GB SSD, I never set up swap, and over the course of four years or so this only bit me about three times, and I spent a lot of that time running two substantial Firefox sessions, a 2GB RAM VirtualBox VM and a handful of other things like urxvt, Inkscape and Chrome all at once.

What I love about linux is that we have the option of doing this. My main machine in my office is from 2009- 4GB RAM, linux mint and fluxbox. It's time to upgrade as I've gotten away a bit from slinging 1s and 0s but damn- still does a job for that.

Similar experience for me, tried to enjoy Ubuntu but Mint comes up more polished and, especially when using the dedicated XFCE release, much faster than default Ubuntu. Still worth doing the basics in terms of stopping services not needed for desktop/laptop use. I ended up removing 8 (!) resident services that were all of no use to me - tadaa, more RAM and responsiveness gained.

Note, I used: "systemctl list-unit-files --type=service | grep enabled" to find all enabled services easily and then: "systemd-analyze" for basic timing analysis

Running LXLE (essentially tweaked Lubuntu) exactly on that, and it FLIES. A SSD disk also helps a lot.

From maybe 2015 to 2017 I used an i3 (I forget the clockspeed) dual core processor with 4 gigs of ram daily working as a web developer. It wasn't even on an ssd. It depends on the type of work but for what I was doing it was fast enough.

Granted I was using Mint with the Cinnamon desktop but i'm sure Gnome with Ubuntu would have been fine too.

I installed Windows 10 on it for a few days to try it out and the difference was night and day. The laptop would pause for about a second every maybe 15 seconds or so, which wasn't a thing on Linux.

Linux works fine on that little RAM, but if it's an older machine, you should check out the price to upgrade. Spending $30 on 8GB would be a huge quality of life improvement.

At 4gb, I'd also look at lubuntu. The UI uses less RAM than Gnome and at that amount of RAM, 1-200MB is significant (plus, LXDE uses less CPU power too). I actually run i3 on a lot of older systems to save even more (but you'll have to learn how to using tiling systems -- not hard, but something more to learn nonetheless).

Does this apply to computers that do not have Intel Xeon Scalable processors? Because if not, that's a glaring omission from the title

This is again deep linked to a specific test page.

Yes, specifically the page with the overview.

They probably just stopped including Amazon referral malware.

Reply on news.ycombinator.com