Comment 26 for bug 1833322

Hi Christian,

Thank you. Yes I was not arguing strictly against irqbalance, just trying
to ascertain some discussion parameters as well as parameters for data
collection.

I have not yet seen a coherent philosophy on what it means to "optimize
performance" with default settings that serve the greatest capacity of
server or desktop scenarios. In my humble opinion, data collection is
useless without this framework of understanding what it is we are trying to
achieve and why in terms of system performance. To me this is the deeper
unresolved issue, perhaps.

I fear that systems are currently optimized by default for throughput. For
users, responsiveness (which can include but is not limited to throughput)
and latency may be more important psychologically (there is an analogy to
this in AV production: we can actually get by with fairly poor video
quality--which consumes the most bandwidth and processor power--if audio
quality remains adequate; ie, audio quality has a disproportionately high
impact on psychology compared to video, especially per unit of data or
processing power allocated). And power saving is important in global terms,
as even small gains multiplied over hundreds or thousands of deployments
can have a significant impact, even if the client or operator doesn't
notice much.

ethan

On Wed, Jan 10, 2024 at 4:35 AM Christian Ehrhardt  <
<email address hidden>> wrote:

> Hi Ethanay
> > All I can find is a recommendation not to use it on CPUs with 2 or fewer
> > cores as the overhead is said to be too high
>
> This isn't a real problem anyway, the service will stop immediately if only
> running on one core - even if running on multiple cores with the same
> cache (as the intended benefit is due to cache hotness by having all I/O
> hitting the same cache).
>
> > I can imagine it might still add undesirable or even critical latency in
> > applications that are highly latency sensitive
>
> I understand your line of thought, but it might even improve latency.
> If there is no bottleneck on the cores assigned to handle an IRQ then
> the improved cache hit rate will make even latency better.
> And if there is a strong bottleneck, then some drivers without IRQbalance
> would end up locked on one cpu - so again these might gain lower latency.
> But I have no data on this either (just like no one seems to have on almost
> any of this).
>
> Just like others I'd personally more expect the drawback to be on a
> potential
> lack of power saving.
>
> > This website gave me some clarity on the theory and purpose:
> > https://www.baeldung.com/linux/irqbalance-modern-hardware
>
> Hah, didn't find this one yet - thank you!
> But to me it only underlines the "it can help as much or even more often"
> expectation.
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1833322
>
> Title:
> Consider removing irqbalance from default install on desktop images
>
> Status in irqbalance package in Ubuntu:
> Confirmed
> Status in ubuntu-meta package in Ubuntu:
> Confirmed
>
> Bug description:
> as per https://github.com/pop-os/default-settings/issues/60
>
> Distribution (run cat /etc/os-release):
>
> $ cat /etc/os-release
> NAME="Pop!_OS"
> VERSION="19.04"
> ID=ubuntu
> ID_LIKE=debian
> PRETTY_NAME="Pop!_OS 19.04"
> VERSION_ID="19.04"
> HOME_URL="https://system76.com/pop"
> SUPPORT_URL="http://support.system76.com"
> BUG_REPORT_URL="https://github.com/pop-os/pop/issues"
> PRIVACY_POLICY_URL="https://system76.com/privacy"
> VERSION_CODENAME=disco
> UBUNTU_CODENAME=disco
>
> Related Application and/or Package Version (run apt policy $PACKAGE
> NAME):
>
> $ apt policy irqbalance
> irqbalance:
> Installed: 1.5.0-3ubuntu1
> Candidate: 1.5.0-3ubuntu1
> Version table:
> *** 1.5.0-3ubuntu1 500
> 500 http://us.archive.ubuntu.com/ubuntu disco/main amd64 Packages
> 100 /var/lib/dpkg/status
>
> $ apt rdepends irqbalance
> irqbalance
> Reverse Depends:
> Recommends: ubuntu-standard
> gce-compute-image-packages
>
> Issue/Bug Description:
>
> as per konkor/cpufreq#48 and
> http://konkor.github.io/cpufreq/faq/#irqbalance-detected
>
> irqbalance is technically not needed on desktop systems (supposedly it
> is mainly for servers), and may actually reduce performance and power
> savings. It appears to provide benefits only to server environments
> that have relatively-constant loading. If it is truly a server-
> oriented package, then it shouldn't be installed by default on a
> desktop/laptop system and shouldn't be included in desktop OS images.
>
> Steps to reproduce (if you know):
>
> This is potentially an issue with all default installs.
>
> Expected behavior:
>
> n/a
>
> Other Notes:
>
> I can safely remove it via "sudo apt purge irqbalance" without any
> apparent adverse side-effects. If someone is running a situation where
> they need it, then they always have the option of installing it from
> the repositories.
>
> To manage notifications about this bug go to:
>
> https://bugs.launchpad.net/ubuntu/+source/irqbalance/+bug/1833322/+subscriptions
>
>