[opensuse-factory] blk-mq schedulers in Tumbleweed
Hi all, As you know, we've got blk-mq merged to kernel since 4.13, introducing, in particular, BFQ — an multithreaded low-latency desktop-oriented I/O scheduler, hopefully to eliminate the infamous #12309 kernel bug. So, I've tweaked a `60-ssd-scheduler.rules` udev rule (it did set scheduler to `deadline` for SSD drives) and made it support blk-mq as well, should it has been enabled in kernel command line; there it goes: https://github.com/develop7/systemd/blob/8463b5cb40d2bfbadda4406c5d6cb9c585f.... There's a PR featuring this tweak at https://github.com/openSUSE/systemd/pull/7, hopefully it'd be merged. Works well for me. Feedback is welcome. -- Regards, Andrei Dziahel -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
On mardi, 2 janvier 2018 11.10:40 h CET Andrei Dziahel wrote:
Hi all,
As you know, we've got blk-mq merged to kernel since 4.13, introducing, in particular, BFQ — an multithreaded low-latency desktop-oriented I/O scheduler, hopefully to eliminate the infamous #12309 kernel bug.
So, I've tweaked a `60-ssd-scheduler.rules` udev rule (it did set scheduler to `deadline` for SSD drives) and made it support blk-mq as well, should it has been enabled in kernel command line; there it goes: https://github.com/develop7/systemd/blob/8463b5cb40d2bfbadda4406c5d6cb9c585 f64e47/rules/60-io-scheduler.rules. There's a PR featuring this tweak at https://github.com/openSUSE/systemd/pull/7, hopefully it'd be merged.
Works well for me. Feedback is welcome.
What about the support of nvme devices ? I didn't see them in the rules mentionned. -- Bruno Friedmann Ioda-Net Sàrl www.ioda-net.ch Bareos Partner, openSUSE Member, fsfe fellowship GPG KEY : D5C9B751C4653227 irc: tigerfoot -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
On Tuesday 2018-01-02 11:29, Bruno Friedmann wrote:
So, I've tweaked a `60-ssd-scheduler.rules` udev rule (it did set scheduler to `deadline` for SSD drives) and made it support blk-mq as well, should it has been enabled in kernel command line; there it goes: https://github.com/develop7/systemd/blob/8463b5cb40d2bfbadda4406c5d6cb9c585 f64e47/rules/60-io-scheduler.rules. There's a PR featuring this tweak at https://github.com/openSUSE/systemd/pull/7, hopefully it'd be merged.
Works well for me. Feedback is welcome.
What about the support of nvme devices ? I didn't see them in the rules mentionned.
Should be classified as rotational=0. -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
AFAIR NVMes are scheduled with blk-mq unconditionally (the link is
buried in history somewhere), so there's no need to handle them nor
here nor at all.
2018-01-02 13:29 GMT+03:00 Bruno Friedmann
On mardi, 2 janvier 2018 11.10:40 h CET Andrei Dziahel wrote:
Hi all,
As you know, we've got blk-mq merged to kernel since 4.13, introducing, in particular, BFQ — an multithreaded low-latency desktop-oriented I/O scheduler, hopefully to eliminate the infamous #12309 kernel bug.
So, I've tweaked a `60-ssd-scheduler.rules` udev rule (it did set scheduler to `deadline` for SSD drives) and made it support blk-mq as well, should it has been enabled in kernel command line; there it goes: https://github.com/develop7/systemd/blob/8463b5cb40d2bfbadda4406c5d6cb9c585 f64e47/rules/60-io-scheduler.rules. There's a PR featuring this tweak at https://github.com/openSUSE/systemd/pull/7, hopefully it'd be merged.
Works well for me. Feedback is welcome.
What about the support of nvme devices ? I didn't see them in the rules mentionned.
--
Bruno Friedmann Ioda-Net Sàrl www.ioda-net.ch Bareos Partner, openSUSE Member, fsfe fellowship GPG KEY : D5C9B751C4653227 irc: tigerfoot
-- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
-- Regards, Andrei Dziahel -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
02.01.2018 13:29, Bruno Friedmann пишет:
On mardi, 2 janvier 2018 11.10:40 h CET Andrei Dziahel wrote:
Hi all,
As you know, we've got blk-mq merged to kernel since 4.13, introducing, in particular, BFQ — an multithreaded low-latency desktop-oriented I/O scheduler, hopefully to eliminate the infamous #12309 kernel bug.
So, I've tweaked a `60-ssd-scheduler.rules` udev rule (it did set scheduler to `deadline` for SSD drives) and made it support blk-mq as well, should it has been enabled in kernel command line; there it goes: https://github.com/develop7/systemd/blob/8463b5cb40d2bfbadda4406c5d6cb9c585 f64e47/rules/60-io-scheduler.rules. There's a PR featuring this tweak at https://github.com/openSUSE/systemd/pull/7, hopefully it'd be merged.
Works well for me. Feedback is welcome.
What about the support of nvme devices ? I didn't see them in the rules mentionned.
NVMe devices do not even use IO scheduler ... -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
On Tue, 02 Jan 2018 11:10:40 +0100, Andrei Dziahel wrote:
Hi all,
As you know, we've got blk-mq merged to kernel since 4.13, introducing, in particular, BFQ — an multithreaded low-latency desktop-oriented I/O scheduler, hopefully to eliminate the infamous #12309 kernel bug.
So, I've tweaked a `60-ssd-scheduler.rules` udev rule (it did set scheduler to `deadline` for SSD drives) and made it support blk-mq as well, should it has been enabled in kernel command line; there it goes: https://github.com/develop7/systemd/blob/8463b5cb40d2bfbadda4406c5d6cb9c585f.... There's a PR featuring this tweak at https://github.com/openSUSE/systemd/pull/7, hopefully it'd be merged.
Works well for me. Feedback is welcome.
Some our storage specialists have been evaluating the I/O scheduler choice. Now Cc'ed, so let's wait for their inputs. BTW, the check of cmdline isn't good; it's not always correct e.g. when it's passed via modprobe.d/*.conf. If any, you can check the content of /sys/module/scsi_mod/parameters/use_blk_mq file instead. thanks, Takashi -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
Yeah, checking cmdline isn't reliable for sure, and I've actually ran
into the file you mentioned, I just wasn't sure it's there for all
times. Anyway, I don't know how do I expose its' content to udev, any
ideas?
TIA.
Regards, Andrei.
2018-01-02 16:15 GMT+03:00 Takashi Iwai
On Tue, 02 Jan 2018 11:10:40 +0100, Andrei Dziahel wrote:
Hi all,
As you know, we've got blk-mq merged to kernel since 4.13, introducing, in particular, BFQ — an multithreaded low-latency desktop-oriented I/O scheduler, hopefully to eliminate the infamous #12309 kernel bug.
So, I've tweaked a `60-ssd-scheduler.rules` udev rule (it did set scheduler to `deadline` for SSD drives) and made it support blk-mq as well, should it has been enabled in kernel command line; there it goes: https://github.com/develop7/systemd/blob/8463b5cb40d2bfbadda4406c5d6cb9c585f.... There's a PR featuring this tweak at https://github.com/openSUSE/systemd/pull/7, hopefully it'd be merged.
Works well for me. Feedback is welcome.
Some our storage specialists have been evaluating the I/O scheduler choice. Now Cc'ed, so let's wait for their inputs.
BTW, the check of cmdline isn't good; it's not always correct e.g. when it's passed via modprobe.d/*.conf. If any, you can check the content of /sys/module/scsi_mod/parameters/use_blk_mq file instead.
thanks,
Takashi
-- Regards, Andrei Dziahel -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org
participants (5)
-
Andrei Borzenkov
-
Andrei Dziahel
-
Bruno Friedmann
-
Jan Engelhardt
-
Takashi Iwai