[prev in list] [next in list] [prev in thread] [next in thread]
List: kde-frameworks-devel
Subject: Re: Review Request 130084: Add a pair of flags forcing fsync during copy loop
From: Oswald Buddenhagen <ossi () kde ! org>
Date: 2017-04-15 16:25:46
Message-ID: 20170415162546.4898.34461 () mimi ! kde ! org
[Download RAW message or body]
--===============8760350552427579087==
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
> On April 15, 2017, 10:30 a.m., David Faure wrote:
> > But doesn't this make copying much slower in the normal case? (copying onto a \
> > non-removable harddisk partition).
> > It sounds to me like this should be
> > 1) done internally in kio_file (no Job API for this)
> > 2) only when the destination is a removable device
i've read multiple times that fsync isn't a performance problem on modern file \
systems any more. whatever that may mean. limiting this to cross-device isn't really \
sensible imo - a) one can have multiple internal disks and b) even if the disk stays \
in, at some point the flushing will commence and will be a major slowdown for \
subsequent operations. in fact, this problem is bad enough that the linux kernel \
community realized it (which, in the area of disk i/o, never ceases to amaze) - \
https://lwn.net/Articles/682582/ (obvious followup question: what kernel do you use? \
this code seems to have landed in 4.10)
- Oswald
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/130084/#review103048
-----------------------------------------------------------
On April 15, 2017, 10:28 a.m., KJ Tsanaktsidis wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/130084/
> -----------------------------------------------------------
>
> (Updated April 15, 2017, 10:28 a.m.)
>
>
> Review request for KDE Frameworks, Oswald Buddenhagen and Thiago Macieira.
>
>
> Repository: kio
>
>
> Description
> -------
>
> When copying a large-ish file (~1-2GB) from very fast storage to very slow storage \
> (e.g. an NVME SSD to a cheap USB flash drive) on a machine with lots of RAM, \
> Dolphin displays a progress bar which finishes in a fraction of a second (i.e. as \
> fast as it takes to read the source file into the Linux page cache). Unmounting the \
> drive then of course takes a long time, with only an indeterminate spinner.
> This patch adds an option to force fsync during copy jobs, so that the copy \
> progress bar measures how long it will take to actually copy the file to the \
> destination.
> I've added two flags - Fsync and FsyncCrossFilesystem - to the JobEnum flag. The \
> former will cause all copy operations to fsync during the copy loop, whilst the \
> latter will only fsync copies that are across different filesystems.
> If this patch gets OK'd, I have another patch which adds support for this into the \
> appropriate places in Dolphin. I would think that at least FsyncCrossFilesystem \
> should be the default, but Fsync always might be a little heavy handed. At the \
> least fsync'ing cross-filesystem copies ensures that the unmount won't take \
> forever.
>
> Diffs
> -----
>
> src/core/copyjob.cpp 7c02cb50d7e9c11bbcd9264832357f3fd6dc8c16
> src/core/filecopyjob.cpp 301b7039158b7dc537b9004c14845b3d1d60f8eb
> src/core/job_base.h 0be9629f42277afc5f72d00d0cae5c9c1cd2b8bc
> src/core/slavebase.cpp 3778df813b8568657a2cbd9412c1244f94696a0c
> src/ioslaves/file/file_unix.cpp 3c1b9927e3dd2d0134f77caec6e6b24a0356d26f
>
> Diff: https://git.reviewboard.kde.org/r/130084/diff/
>
>
> Testing
> -------
>
> Tested the patch with KDE/Dolphin on Arch Linux, which is version 5.32.0. The diff \
> applies cleanly to master so I assume there shouldn't be any issues there, but I've \
> not actually checked that. As advertised, copying a file to USB flash storage now \
> displays an accurate progress bar.
> I experimented with how often fsync should be called on my hardware, and I found \
> calling it every ~1M copied caused no decrease in copy performance whilst still \
> providing accurate progress info. That is the setting I've gone with in this patch. \
> I'm open to suggestions on how this could be tuned better though.
>
> Thanks,
>
> KJ Tsanaktsidis
>
>
--===============8760350552427579087==
MIME-Version: 1.0
Content-Type: text/html; charset="utf-8"
Content-Transfer-Encoding: 7bit
<html>
<body>
<div style="font-family: Verdana, Arial, Helvetica, Sans-Serif;">
<table bgcolor="#f9f3c9" width="100%" cellpadding="12" style="border: 1px #c9c399 \
solid; border-radius: 6px; -moz-border-radius: 6px; -webkit-border-radius: 6px;"> \
<tr> <td>
This is an automatically generated e-mail. To reply, visit:
<a href="https://git.reviewboard.kde.org/r/130084/">https://git.reviewboard.kde.org/r/130084/</a>
</td>
</tr>
</table>
<br />
<blockquote style="margin-left: 1em; border-left: 2px solid #d0d0d0; padding-left: \
10px;"> <p style="margin-top: 0;">On April 15th, 2017, 10:30 a.m. CEST, <b>David \
Faure</b> wrote:</p> <blockquote style="margin-left: 1em; border-left: 2px solid \
#d0d0d0; padding-left: 10px;"> <pre style="white-space: pre-wrap; white-space: \
-moz-pre-wrap; white-space: -pre-wrap; white-space: -o-pre-wrap; word-wrap: \
break-word;"><p style="padding: 0;text-rendering: inherit;margin: 0;line-height: \
inherit;white-space: inherit;">But doesn't this make copying much slower in the \
normal case? (copying onto a non-removable harddisk partition).</p> <p \
style="padding: 0;text-rendering: inherit;margin: 0;line-height: inherit;white-space: \
inherit;">It sounds to me like this should be 1) done internally in kio_file (no Job \
API for this) 2) only when the destination is a removable device</p></pre>
</blockquote>
</blockquote>
<pre style="white-space: pre-wrap; white-space: -moz-pre-wrap; white-space: \
-pre-wrap; white-space: -o-pre-wrap; word-wrap: break-word;"><p style="padding: \
0;text-rendering: inherit;margin: 0;line-height: inherit;white-space: inherit;">i've \
read multiple times that fsync isn't a performance problem on modern file systems any \
more. whatever that may mean. limiting this to cross-device isn't really sensible imo \
- a) one can have multiple internal disks and b) even if the disk stays in, at some \
point the flushing will commence and will be a major slowdown for subsequent \
operations. in fact, this problem is bad enough that the linux kernel community \
realized it (which, in the area of disk i/o, never ceases to amaze) - \
https://lwn.net/Articles/682582/ (obvious followup question: what kernel do you use? \
this code seems to have landed in 4.10)</p></pre> <br />
<p>- Oswald</p>
<br />
<p>On April 15th, 2017, 10:28 a.m. CEST, KJ Tsanaktsidis wrote:</p>
<table bgcolor="#fefadf" width="100%" cellspacing="0" cellpadding="12" style="border: \
1px #888a85 solid; border-radius: 6px; -moz-border-radius: 6px; \
-webkit-border-radius: 6px;"> <tr>
<td>
<div>Review request for KDE Frameworks, Oswald Buddenhagen and Thiago Macieira.</div>
<div>By KJ Tsanaktsidis.</div>
<p style="color: grey;"><i>Updated April 15, 2017, 10:28 a.m.</i></p>
<div style="margin-top: 1.5em;">
<b style="color: #575012; font-size: 10pt;">Repository: </b>
kio
</div>
<h1 style="color: #575012; font-size: 10pt; margin-top: 1.5em;">Description </h1>
<table width="100%" bgcolor="#ffffff" cellspacing="0" cellpadding="10" \
style="border: 1px solid #b8b5a0"> <tr>
<td>
<pre style="margin: 0; padding: 0; white-space: pre-wrap; white-space: \
-moz-pre-wrap; white-space: -pre-wrap; white-space: -o-pre-wrap; word-wrap: \
break-word;"><p style="padding: 0;text-rendering: inherit;margin: 0;line-height: \
inherit;white-space: inherit;">When copying a large-ish file (~1-2GB) from very fast \
storage to very slow storage (e.g. an NVME SSD to a cheap USB flash drive) on a \
machine with lots of RAM, Dolphin displays a progress bar which finishes in a \
fraction of a second (i.e. as fast as it takes to read the source file into the Linux \
page cache). Unmounting the drive then of course takes a long time, with only an \
indeterminate spinner.</p> <p style="padding: 0;text-rendering: inherit;margin: \
0;line-height: inherit;white-space: inherit;">This patch adds an option to force \
fsync during copy jobs, so that the copy progress bar measures how long it will take \
to actually copy the file to the destination.</p> <p style="padding: \
0;text-rendering: inherit;margin: 0;line-height: inherit;white-space: inherit;">I've \
added two flags - Fsync and FsyncCrossFilesystem - to the JobEnum flag. The former \
will cause all copy operations to fsync during the copy loop, whilst the latter will \
only fsync copies that are across different filesystems.</p> <p style="padding: \
0;text-rendering: inherit;margin: 0;line-height: inherit;white-space: inherit;">If \
this patch gets OK'd, I have another patch which adds support for this into the \
appropriate places in Dolphin. I would think that at least FsyncCrossFilesystem \
should be the default, but Fsync always might be a little heavy handed. At the least \
fsync'ing cross-filesystem copies ensures that the unmount won't take \
forever.</p></pre> </td>
</tr>
</table>
<h1 style="color: #575012; font-size: 10pt; margin-top: 1.5em;">Testing </h1>
<table width="100%" bgcolor="#ffffff" cellspacing="0" cellpadding="10" style="border: \
1px solid #b8b5a0"> <tr>
<td>
<pre style="margin: 0; padding: 0; white-space: pre-wrap; white-space: \
-moz-pre-wrap; white-space: -pre-wrap; white-space: -o-pre-wrap; word-wrap: \
break-word;"><p style="padding: 0;text-rendering: inherit;margin: 0;line-height: \
inherit;white-space: inherit;">Tested the patch with KDE/Dolphin on Arch Linux, which \
is version 5.32.0. The diff applies cleanly to master so I assume there shouldn't be \
any issues there, but I've not actually checked that. As advertised, copying a file \
to USB flash storage now displays an accurate progress bar.</p> <p style="padding: \
0;text-rendering: inherit;margin: 0;line-height: inherit;white-space: inherit;">I \
experimented with how often fsync should be called on my hardware, and I found \
calling it every ~1M copied caused no decrease in copy performance whilst still \
providing accurate progress info. That is the setting I've gone with in this patch. \
I'm open to suggestions on how this could be tuned better though.</p></pre> </td>
</tr>
</table>
<h1 style="color: #575012; font-size: 10pt; margin-top: 1.5em;">Diffs</b> </h1>
<ul style="margin-left: 3em; padding-left: 0;">
<li>src/core/copyjob.cpp <span style="color: \
grey">(7c02cb50d7e9c11bbcd9264832357f3fd6dc8c16)</span></li>
<li>src/core/filecopyjob.cpp <span style="color: \
grey">(301b7039158b7dc537b9004c14845b3d1d60f8eb)</span></li>
<li>src/core/job_base.h <span style="color: \
grey">(0be9629f42277afc5f72d00d0cae5c9c1cd2b8bc)</span></li>
<li>src/core/slavebase.cpp <span style="color: \
grey">(3778df813b8568657a2cbd9412c1244f94696a0c)</span></li>
<li>src/ioslaves/file/file_unix.cpp <span style="color: \
grey">(3c1b9927e3dd2d0134f77caec6e6b24a0356d26f)</span></li>
</ul>
<p><a href="https://git.reviewboard.kde.org/r/130084/diff/" style="margin-left: \
3em;">View Diff</a></p>
</td>
</tr>
</table>
</div>
</body>
</html>
--===============8760350552427579087==--
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic