CFQ I/O priorities only for reads?

CFQ I/O priorities only for reads?

[Martin Steigerwald]

Hi jens und Vivek,

Vivek, I cc'd you, cause you wrote the new cfq-iosched.txt.


In trying to understand how I/O priorities actually really work, I tried to dd 
with

rm nullen-id ; sync ; /usr/bin/time ionice -c3 dd if=/dev/zero of=nullen-id 
count=500 bs=1M conv=fsync

versus

rm nullen-rl; sync ; /usr/bin/time ionice -c1 -n0 dd if=/dev/zero of=nullen-rl 
count=500 bs=1M conv=fsync

concurrently. No differences. At first I was puzzled, then I thought maybe 
direct I/O makes a difference. So I tried with oflag=direct.

And it does.

Then I actually read the documentation block/ioprio.txt (3.1 here):

> With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic io
> priorities are supported for reads on files.  This enables users to io nice
> processes or process groups, similar to what has been possible with cpu
> scheduling for ages.  This document mainly details the current
> possibilities with cfq; other io schedulers do not support io priorities
> thus far.

According to it I/O priorities will even only work on reads. Is that correct? 
I mean they do work on reads, I tested it, but *only* on reads?

>From what I see here, it also works for direct I/O write requests

So from what I conclude is that CFQ I/O priorities work for all requests that 
are issued via synchronous system calls, but not for those issued via 
asynchronous calls, i. e. everything that goes through the pagecache.

Is that correct?


Vivek, one thing on cfq-iosched.txt: Could slice_idle=0 make sense on SSDs? 
Later on you write that there are some SSD optimizations in place that cut 
down idling already.


Thanks,
-- 
Martin Steigerwald - teamix GmbH - http://www.teamix.de
gpg: 19E3 8D42 896F D004 08AC A0CA 1E10 C593 0399 AE90

Re: CFQ I/O priorities only for reads?

[Jens Axboe]

On 2011-11-28 15:42, Martin Steigerwald wrote:
> Hi jens und Vivek,
> 
> Vivek, I cc'd you, cause you wrote the new cfq-iosched.txt.
> 
> 
> In trying to understand how I/O priorities actually really work, I tried to dd 
> with
> 
> rm nullen-id ; sync ; /usr/bin/time ionice -c3 dd if=/dev/zero of=nullen-id 
> count=500 bs=1M conv=fsync
> 
> versus
> 
> rm nullen-rl; sync ; /usr/bin/time ionice -c1 -n0 dd if=/dev/zero of=nullen-rl 
> count=500 bs=1M conv=fsync
> 
> concurrently. No differences. At first I was puzzled, then I thought maybe 
> direct I/O makes a difference. So I tried with oflag=direct.
> 
> And it does.
> 
> Then I actually read the documentation block/ioprio.txt (3.1 here):
> 
>> With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic io
>> priorities are supported for reads on files.  This enables users to io nice
>> processes or process groups, similar to what has been possible with cpu
>> scheduling for ages.  This document mainly details the current
>> possibilities with cfq; other io schedulers do not support io priorities
>> thus far.
> 
> According to it I/O priorities will even only work on reads. Is that correct? 
> I mean they do work on reads, I tested it, but *only* on reads?
> 
> From what I see here, it also works for direct I/O write requests
> 
> So from what I conclude is that CFQ I/O priorities work for all requests that 
> are issued via synchronous system calls, but not for those issued via 
> asynchronous calls, i. e. everything that goes through the pagecache.
> 
> Is that correct?

Priorities work for reads AND direct writes. In other words, it does not
work for buffered writes.

> Vivek, one thing on cfq-iosched.txt: Could slice_idle=0 make sense on
> SSDs?  Later on you write that there are some SSD optimizations in
> place that cut down idling already.

It will have a functional difference even on SSDs, depending on your
workload, even if the scope of idling is smaller on an SSD.

-- 
Jens Axboe

[PATCH 1/1] Mention that I/O priorities also work on direct writes.

[Martin Steigerwald]

Am Montag, 28. November 2011 schrieb Jens Axboe:
> On 2011-11-28 15:42, Martin Steigerwald wrote:
> > Hi jens und Vivek,
> > 
> > Vivek, I cc'd you, cause you wrote the new cfq-iosched.txt.
> > 
> > 
> > In trying to understand how I/O priorities actually really work, I tried
> > to dd with
> > 
> > rm nullen-id ; sync ; /usr/bin/time ionice -c3 dd if=/dev/zero
> > of=nullen-id count=500 bs=1M conv=fsync
> > 
> > versus
> > 
> > rm nullen-rl; sync ; /usr/bin/time ionice -c1 -n0 dd if=/dev/zero
> > of=nullen-rl count=500 bs=1M conv=fsync
> > 
> > concurrently. No differences. At first I was puzzled, then I thought
> > maybe direct I/O makes a difference. So I tried with oflag=direct.
> > 
> > And it does.
> > 
> > Then I actually read the documentation block/ioprio.txt (3.1 here):
> >> With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic
> >> io priorities are supported for reads on files.  This enables users to
> >> io nice processes or process groups, similar to what has been possible
> >> with cpu scheduling for ages.  This document mainly details the current
> >> possibilities with cfq; other io schedulers do not support io priorities
> >> thus far.
> > 
> > According to it I/O priorities will even only work on reads. Is that
> > correct? I mean they do work on reads, I tested it, but *only* on reads?
> > 
> > From what I see here, it also works for direct I/O write requests
> > 
> > So from what I conclude is that CFQ I/O priorities work for all requests
> > that are issued via synchronous system calls, but not for those issued
> > via asynchronous calls, i. e. everything that goes through the
> > pagecache.
> > 
> > Is that correct?
> 
> Priorities work for reads AND direct writes. In other words, it does not
> work for buffered writes.
> 
> > Vivek, one thing on cfq-iosched.txt: Could slice_idle=0 make sense on
> > SSDs?  Later on you write that there are some SSD optimizations in
> > place that cut down idling already.
> 
> It will have a functional difference even on SSDs, depending on your
> workload, even if the scope of idling is smaller on an SSD.

>From 5414ce9fd8c384a3a25a478490a022539694e4e0 Mon Sep 17 00:00:00 2001
From: Martin Steigerwald <ms@teamix.de>
Date: Mon, 28 Nov 2011 16:10:32 +0100
Subject: [PATCH 1/3] Mention that I/O priorities also work on direct writes.

---
 Documentation/block/ioprio.txt |    9 +++++----
 1 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/Documentation/block/ioprio.txt b/Documentation/block/ioprio.txt
index 8ed8c59..a555c59 100644
--- a/Documentation/block/ioprio.txt
+++ b/Documentation/block/ioprio.txt
@@ -6,10 +6,11 @@ Intro
 -----
 
 With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic io
-priorities are supported for reads on files.  This enables users to io nice
-processes or process groups, similar to what has been possible with cpu
-scheduling for ages.  This document mainly details the current possibilities
-with cfq; other io schedulers do not support io priorities thus far.
+priorities are supported for reads and direct, not buffered, writes on files
+This enables users to io nice processes or process groups, similar to what
+has been possible with cpu scheduling for ages. This document mainly details
+the current possibilities with cfq; other io schedulers do not support io
+priorities thus far.
 
 Scheduling classes
 ------------------
-- 
1.7.7.3

Thanks,
-- 
Martin Steigerwald - teamix GmbH - http://www.teamix.de
gpg: 19E3 8D42 896F D004 08AC A0CA 1E10 C593 0399 AE90

[PATCH 2/3] Replace io by I/O where approbiate.

[Martin Steigerwald]

Am Montag, 28. November 2011 schrieb Jens Axboe:
> On 2011-11-28 15:42, Martin Steigerwald wrote:
> > Hi jens und Vivek,
> > 
> > Vivek, I cc'd you, cause you wrote the new cfq-iosched.txt.
> > 
> > 
> > In trying to understand how I/O priorities actually really work, I tried
> > to dd with
> > 
> > rm nullen-id ; sync ; /usr/bin/time ionice -c3 dd if=/dev/zero
> > of=nullen-id count=500 bs=1M conv=fsync
> > 
> > versus
> > 
> > rm nullen-rl; sync ; /usr/bin/time ionice -c1 -n0 dd if=/dev/zero
> > of=nullen-rl count=500 bs=1M conv=fsync
> > 
> > concurrently. No differences. At first I was puzzled, then I thought
> > maybe direct I/O makes a difference. So I tried with oflag=direct.
> > 
> > And it does.
> > 
> > Then I actually read the documentation block/ioprio.txt (3.1 here):
> >> With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic
> >> io priorities are supported for reads on files.  This enables users to
> >> io nice processes or process groups, similar to what has been possible
> >> with cpu scheduling for ages.  This document mainly details the current
> >> possibilities with cfq; other io schedulers do not support io priorities
> >> thus far.
> > 
> > According to it I/O priorities will even only work on reads. Is that
> > correct? I mean they do work on reads, I tested it, but *only* on reads?
> > 
> > From what I see here, it also works for direct I/O write requests
> > 
> > So from what I conclude is that CFQ I/O priorities work for all requests
> > that are issued via synchronous system calls, but not for those issued
> > via asynchronous calls, i. e. everything that goes through the
> > pagecache.
> > 
> > Is that correct?
> 
> Priorities work for reads AND direct writes. In other words, it does not
> work for buffered writes.

I also replaced io by I/O where approbiate:

>From 10a319a1c1782e40c84fddb79f0f7ed719ece774 Mon Sep 17 00:00:00 2001
From: Martin Steigerwald <ms@teamix.de>
Date: Mon, 28 Nov 2011 16:12:55 +0100
Subject: [PATCH 2/3] Replace io by I/O where approbiate.

---
 Documentation/block/ioprio.txt |   20 ++++++++++----------
 1 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/Documentation/block/ioprio.txt b/Documentation/block/ioprio.txt
index a555c59..4775a95 100644
--- a/Documentation/block/ioprio.txt
+++ b/Documentation/block/ioprio.txt
@@ -5,37 +5,37 @@ Block io priorities
 Intro
 -----
 
-With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic io
+With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic I/O
 priorities are supported for reads and direct, not buffered, writes on files
-This enables users to io nice processes or process groups, similar to what
+This enables users to I/O nice processes or process groups, similar to what
 has been possible with cpu scheduling for ages. This document mainly details
-the current possibilities with cfq; other io schedulers do not support io
+the current possibilities with cfq; other I/O schedulers do not support I/O
 priorities thus far.
 
 Scheduling classes
 ------------------
 
-CFQ implements three generic scheduling classes that determine how io is
+CFQ implements three generic scheduling classes that determine how I/O is
 served for a process.
 
-IOPRIO_CLASS_RT: This is the realtime io class. This scheduling class is given
+IOPRIO_CLASS_RT: This is the realtime I/O class. This scheduling class is given
 higher priority than any other in the system, processes from this class are
 given first access to the disk every time. Thus it needs to be used with some
-care, one io RT process can starve the entire system. Within the RT class,
+care, one I/O RT process can starve the entire system. Within the RT class,
 there are 8 levels of class data that determine exactly how much time this
 process needs the disk for on each service. In the future this might change
 to be more directly mappable to performance, by passing in a wanted data
 rate instead.
 
 IOPRIO_CLASS_BE: This is the best-effort scheduling class, which is the default
-for any process that hasn't set a specific io priority. The class data
-determines how much io bandwidth the process will get, it's directly mappable
+for any process that hasn't set a specific I/O priority. The class data
+determines how much I/O bandwidth the process will get, it's directly mappable
 to the cpu nice levels just more coarsely implemented. 0 is the highest
-BE prio level, 7 is the lowest. The mapping between cpu nice level and io
+BE prio level, 7 is the lowest. The mapping between cpu nice level and I/O
 nice level is determined as: io_nice = (cpu_nice + 20) / 5.
 
 IOPRIO_CLASS_IDLE: This is the idle scheduling class, processes running at this
-level only get io time when no one else needs the disk. The idle class has no
+level only get I/O time when no one else needs the disk. The idle class has no
 class data, since it doesn't really apply here.
 
 Tools
-- 
1.7.7.3


Thanks,
-- 
Martin Steigerwald - teamix GmbH - http://www.teamix.de
gpg: 19E3 8D42 896F D004 08AC A0CA 1E10 C593 0399 AE90

[PATCH 3/3] Mention that the util-linux package provides an ionice command.

[Martin Steigerwald]

Am Montag, 28. November 2011 schrieb Martin Steigerwald:
> Hi jens und Vivek,
> 
> Vivek, I cc'd you, cause you wrote the new cfq-iosched.txt.
> 
> 
> In trying to understand how I/O priorities actually really work, I tried to
> dd with
> 
> rm nullen-id ; sync ; /usr/bin/time ionice -c3 dd if=/dev/zero of=nullen-id
> count=500 bs=1M conv=fsync
> 
> versus
> 
> rm nullen-rl; sync ; /usr/bin/time ionice -c1 -n0 dd if=/dev/zero
> of=nullen-rl count=500 bs=1M conv=fsync
> 
> concurrently. No differences. At first I was puzzled, then I thought maybe
> direct I/O makes a difference. So I tried with oflag=direct.
> 
> And it does.
> 
> Then I actually read the documentation block/ioprio.txt (3.1 here):
> > With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic io
> > priorities are supported for reads on files.  This enables users to io
> > nice processes or process groups, similar to what has been possible with
> > cpu scheduling for ages.  This document mainly details the current
> > possibilities with cfq; other io schedulers do not support io priorities
> > thus far.
> 
> According to it I/O priorities will even only work on reads. Is that
> correct? I mean they do work on reads, I tested it, but *only* on reads?
> 
> From what I see here, it also works for direct I/O write requests
> 
> So from what I conclude is that CFQ I/O priorities work for all requests
> that are issued via synchronous system calls, but not for those issued via
> asynchronous calls, i. e. everything that goes through the pagecache.
> 
> Is that correct?
> 
> 
> Vivek, one thing on cfq-iosched.txt: Could slice_idle=0 make sense on SSDs?
> Later on you write that there are some SSD optimizations in place that cut
> down idling already.

And mentioned that ionice is available from the util-linux package:


>From 182d2e06d65ea784c2bbf2e756ce0452fdebef8c Mon Sep 17 00:00:00 2001
From: Martin Steigerwald <ms@teamix.de>
Date: Mon, 28 Nov 2011 16:16:33 +0100
Subject: [PATCH 3/3] Mention that the util-linux package provides an ionice
 command.

---
 Documentation/block/ioprio.txt |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/Documentation/block/ioprio.txt b/Documentation/block/ioprio.txt
index 4775a95..5b79a29 100644
--- a/Documentation/block/ioprio.txt
+++ b/Documentation/block/ioprio.txt
@@ -58,6 +58,9 @@ For a running process, you can give the pid instead:
 
 will change pid 100 to run at the realtime scheduling class, at priority 2.
 
+The util-linux package includes an ionice command that basically works like
+described here.
+
 ---> snip ionice.c tool <---
 
 #include <stdio.h>
-- 
1.7.7.3

Thanks,
-- 
Martin Steigerwald - teamix GmbH - http://www.teamix.de
gpg: 19E3 8D42 896F D004 08AC A0CA 1E10 C593 0399 AE90

Re: [PATCH 1/1] Mention that I/O priorities also work on direct writes.

[Martin Steigerwald]

Hi Jens!

Did you notice the three patches I send?

Thanks,
Martin

----- Ursprüngliche Mail -----
> Am Montag, 28. November 2011 schrieb Jens Axboe:
> > On 2011-11-28 15:42, Martin Steigerwald wrote:
> > > Hi jens und Vivek,
> > > 
> > > Vivek, I cc'd you, cause you wrote the new cfq-iosched.txt.
> > > 
> > > 
> > > In trying to understand how I/O priorities actually really work,
> > > I tried
> > > to dd with
> > > 
> > > rm nullen-id ; sync ; /usr/bin/time ionice -c3 dd if=/dev/zero
> > > of=nullen-id count=500 bs=1M conv=fsync
> > > 
> > > versus
> > > 
> > > rm nullen-rl; sync ; /usr/bin/time ionice -c1 -n0 dd if=/dev/zero
> > > of=nullen-rl count=500 bs=1M conv=fsync
> > > 
> > > concurrently. No differences. At first I was puzzled, then I
> > > thought
> > > maybe direct I/O makes a difference. So I tried with
> > > oflag=direct.
> > > 
> > > And it does.
> > > 
> > > Then I actually read the documentation block/ioprio.txt (3.1
> > > here):
> > >> With the introduction of cfq v3 (aka cfq-ts or time sliced cfq),
> > >> basic
> > >> io priorities are supported for reads on files.  This enables
> > >> users to
> > >> io nice processes or process groups, similar to what has been
> > >> possible
> > >> with cpu scheduling for ages.  This document mainly details the
> > >> current
> > >> possibilities with cfq; other io schedulers do not support io
> > >> priorities
> > >> thus far.
> > > 
> > > According to it I/O priorities will even only work on reads. Is
> > > that
> > > correct? I mean they do work on reads, I tested it, but *only* on
> > > reads?
> > > 
> > > From what I see here, it also works for direct I/O write requests
> > > 
> > > So from what I conclude is that CFQ I/O priorities work for all
> > > requests
> > > that are issued via synchronous system calls, but not for those
> > > issued
> > > via asynchronous calls, i. e. everything that goes through the
> > > pagecache.
> > > 
> > > Is that correct?
> > 
> > Priorities work for reads AND direct writes. In other words, it
> > does not
> > work for buffered writes.
> > 
> > > Vivek, one thing on cfq-iosched.txt: Could slice_idle=0 make
> > > sense on
> > > SSDs?  Later on you write that there are some SSD optimizations
> > > in
> > > place that cut down idling already.
> > 
> > It will have a functional difference even on SSDs, depending on
> > your
> > workload, even if the scope of idling is smaller on an SSD.
> 
> From 5414ce9fd8c384a3a25a478490a022539694e4e0 Mon Sep 17 00:00:00
> 2001
> From: Martin Steigerwald <ms@teamix.de>
> Date: Mon, 28 Nov 2011 16:10:32 +0100
> Subject: [PATCH 1/3] Mention that I/O priorities also work on direct
> writes.
> 
> ---
>  Documentation/block/ioprio.txt |    9 +++++----
>  1 files changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/Documentation/block/ioprio.txt
> b/Documentation/block/ioprio.txt
> index 8ed8c59..a555c59 100644
> --- a/Documentation/block/ioprio.txt
> +++ b/Documentation/block/ioprio.txt
> @@ -6,10 +6,11 @@ Intro
>  -----
>  
>  With the introduction of cfq v3 (aka cfq-ts or time sliced cfq),
>  basic io
> -priorities are supported for reads on files.  This enables users to
> io nice
> -processes or process groups, similar to what has been possible with
> cpu
> -scheduling for ages.  This document mainly details the current
> possibilities
> -with cfq; other io schedulers do not support io priorities thus far.
> +priorities are supported for reads and direct, not buffered, writes
> on files
> +This enables users to io nice processes or process groups, similar
> to what
> +has been possible with cpu scheduling for ages. This document mainly
> details
> +the current possibilities with cfq; other io schedulers do not
> support io
> +priorities thus far.
>  
>  Scheduling classes
>  ------------------
> --
> 1.7.7.3
> 
> Thanks,
> --
> Martin Steigerwald - teamix GmbH - http://www.teamix.de
> gpg: 19E3 8D42 896F D004 08AC A0CA 1E10 C593 0399 AE90
> --
> To unsubscribe from this list: send the line "unsubscribe
> linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 

-- 
Martin Steigerwald
Trainer / Consultant

teamix GmbH
Solide IT-Infrastruktur
Südwestpark 35
90449 Nürnberg

fon:   +49 (911) 30999- 0
fax:   +49 (911) 30999-99
mail:  ms@teamix.de
web:   http://www.teamix.de
vcf:   http://www.teamix.de/vcf/ms.vcf
gpg:   19E3 8D42 896F D004 08AC
       A0CA 1E10 C593 0399 AE90

Amtsgericht Nürnberg, HRB 18320
Geschäftsführer: Oliver Kügow, Richard Müller