Can "intense" I/O make a PC barely usable?

0BADC0DE · 2018-12-11 19:05:27

"Yes of course" is a general answer. I do know.

My PC has a 3.40 GHz i7-4770 with 32GB RAM and 2x3TB soft RAID1 for storage.
It's running the latest Linux-Zen kernel.
While I am making a disk image from an USB3 disk to the internal storage, my system is barely usable.
The mouse pointer vanishes, the keystrokes a lagging, anything I try to do is slooow.
I am doing the image with a command like this:

nice -n 19 ionice -c 3 dd status=progress if=/dev/sde of=~/tmp/disk.img bs=4M

The objective is to have a low priority task (nice -n 19) with the I/O done "in the spare time" (ionice -c 3).
The average I/O speed as reported by dd is between 85 MB/s and 110 MB/s.
The hardware I have is:

00:00.0 Host bridge: Intel Corporation 4th Gen Core Processor DRAM Controller (rev 06)
00:02.0 VGA compatible controller: Intel Corporation Xeon E3-1200 v3/4th Gen Core Processor Integrated Graphics Controller (rev 06)
00:14.0 USB controller: Intel Corporation 8 Series/C220 Series Chipset Family USB xHCI (rev 04)
00:16.0 Communication controller: Intel Corporation 8 Series/C220 Series Chipset Family MEI Controller #1 (rev 04)
00:1a.0 USB controller: Intel Corporation 8 Series/C220 Series Chipset Family USB EHCI #2 (rev 04)
00:1b.0 Audio device: Intel Corporation 8 Series/C220 Series Chipset High Definition Audio Controller (rev 04)
00:1c.0 PCI bridge: Intel Corporation 8 Series/C220 Series Chipset Family PCI Express Root Port #1 (rev d4)
00:1c.2 PCI bridge: Intel Corporation 8 Series/C220 Series Chipset Family PCI Express Root Port #3 (rev d4)
00:1c.5 PCI bridge: Intel Corporation 8 Series/C220 Series Chipset Family PCI Express Root Port #6 (rev d4)
00:1d.0 USB controller: Intel Corporation 8 Series/C220 Series Chipset Family USB EHCI #1 (rev 04)
00:1f.0 ISA bridge: Intel Corporation H87 Express LPC Controller (rev 04)
00:1f.2 SATA controller: Intel Corporation 8 Series/C220 Series Chipset Family 6-port SATA Controller 1 [AHCI mode] (rev 04)
00:1f.3 SMBus: Intel Corporation 8 Series/C220 Series Chipset Family SMBus Controller (rev 04)
02:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller (rev 0c)
03:00.0 USB controller: Renesas Technology Corp. uPD720201 USB 3.0 Host Controller (rev 03)

I am not using the "Renesas" USB expansion card but rather the internal USB ports (Intel's) for the USB3 external disk.

I am not discussing the I/O speed here (though it'd be higher than 100 MB/s)
I am arguing about the fact that a single I/O operation (actually 2: a read and a write) can kill an otherwise "nicely powerful" PC.
How can it be possible?
I my opinion, even without a changing the default nice and ionice values, interactive tasks shouldn't be jeopardized by an I/O intensive one.
Especially if the used I/O bandwidth is "just a mere" 100 MB/s (times two).
Even if I explicitly asked for an "idle" I/O scheduling class and the lowest CPU scheduling class.
I'd like to do the same with lower speed but with a usable PC.
Is there a way to do that?
Or it's my hardware that is forcing such a weird behavior?

Last edited by 0BADC0DE (2018-12-11 19:06:39)

jasonwryan · 2018-12-11 19:33:35

Can you reproduce on the stock kernel?

loqs · 2018-12-11 19:57:44

man 1 ionice wrote:

NOTES
Linux supports I/O scheduling priorities and classes since 2.6.13 with
the CFQ I/O scheduler.

0BADC0DE · 2018-12-12 08:52:55

loqs wrote:

man 1 ionice wrote:
NOTES
Linux supports I/O scheduling priorities and classes since 2.6.13 with
the CFQ I/O scheduler.

Thanks for the hint.
But...

~ dmesg | grep -i cfq
[    0.874669] io scheduler cfq registered (default)

Last edited by 0BADC0DE (2018-12-12 08:54:55)

0BADC0DE · 2018-12-12 08:54:33

jasonwryan wrote:

Can you reproduce on the stock kernel?

I could, in theory.
But I'd need to do a reboot, which isn't really possible right now.
Maybe in the next weekend.
Thanks for the feed back.

seth · 2018-12-12 09:29:10

Do you see this effect w/o a display server (+DE/compositor) ie. on just two consoles as well?
Is there unexpected CPU/RAM load?

0BADC0DE · 2018-12-12 09:40:36

seth wrote:

Do you see this effect w/o a display server (+DE/compositor) ie. on just two consoles as well?

Hard to say.
I was trying to do that image while using the PC which is running Plasma/KDE.
I was editing some files with vim and running Chromium to read documentation.
No "heavy" task but that disk dump.
I'd test without graphics, but I need to wait until the weekend: that's my main productivity PC.

seth wrote:

Is there unexpected CPU/RAM load?

Both iotop and htop weren't showing anything unexpected.
Load average was around 1.8 which is rather low with an 8-core CPU like mine.
RAM usage was negligible (about 3 GB RAM).
I suspect a serious bottleneck either with the RAM (it's a dual channel system) or with the global I/O (CPU<-->devices) bandwidth.
And, anyway, I am puzzled by the total ineffectiveness of the "idle" I/O scheduling class!

V1del · 2018-12-12 09:52:34

Afaik CFQ always had problems with that, and as it is being/already is removed in favor of MQ schedulers you might simply try some other scheduler, I did notice similar issues and I did move my main system to BFQ when it became available in the 4.14 kernel which seems to have alleviated most of them, so that might be an avenue to explore.

FWIW that should be runtime changeable and afaik zen has non-MQ BFQ compiled in.

Last edited by V1del (2018-12-12 09:53:55)

0BADC0DE · 2018-12-12 09:59:02

V1del wrote:

Afaik CFQ always had problems with that, and as it is being/already is removed in favor of MQ schedulers you might simply try some other scheduler, I did notice similar issues and I did move my main system to BFQ when it became available in the 4.14 kernel which seems to have alleviated most of them, so that might be an avenue to explore.

I'll give bfq a try later and post my results here.
This is what I have here:

~ dmesg | grep -i scheduler
[    0.874651] io scheduler noop registered
[    0.874652] io scheduler deadline registered
[    0.874669] io scheduler cfq registered (default)
[    0.874670] io scheduler mq-deadline registered
[    0.874671] io scheduler kyber registered
[    0.874684] io scheduler bfq registered
[    1.355592] CFS CPU scheduler.

Thanks.

Last edited by 0BADC0DE (2018-12-12 10:00:04)

loqs · 2018-12-12 10:49:40

0BADC0DE wrote:

~ dmesg | grep -i cfq
[    0.874669] io scheduler cfq registered (default)

That is the default single queue scheduler.
What is the output of

cat /sys/block/sde/queue/scheduler

0BADC0DE · 2018-12-12 10:54:36

loqs wrote:

0BADC0DE wrote:
~ dmesg | grep -i cfq
[    0.874669] io scheduler cfq registered (default)
That is the default single queue scheduler.
What is the output of
cat /sys/block/sde/queue/scheduler

~ cat /sys/block/sde/queue/scheduler
mq-deadline kyber [bfq] none

... interesting!

loqs · 2018-12-12 11:09:16

Improving_performance#Changing_I/O_scheduler

Arch Linux

#1 2018-12-11 19:05:27

Can "intense" I/O make a PC barely usable?

#2 2018-12-11 19:33:35

Re: Can "intense" I/O make a PC barely usable?

#3 2018-12-11 19:57:44

Re: Can "intense" I/O make a PC barely usable?

#4 2018-12-12 08:52:55

Re: Can "intense" I/O make a PC barely usable?

#5 2018-12-12 08:54:33

Re: Can "intense" I/O make a PC barely usable?

#6 2018-12-12 09:29:10

Re: Can "intense" I/O make a PC barely usable?

#7 2018-12-12 09:40:36

Re: Can "intense" I/O make a PC barely usable?

#8 2018-12-12 09:52:34

Re: Can "intense" I/O make a PC barely usable?

#9 2018-12-12 09:59:02

Re: Can "intense" I/O make a PC barely usable?

#10 2018-12-12 10:49:40

Re: Can "intense" I/O make a PC barely usable?

#11 2018-12-12 10:54:36

Re: Can "intense" I/O make a PC barely usable?

#12 2018-12-12 11:09:16

Re: Can "intense" I/O make a PC barely usable?

Board footer