<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"
	xmlns:content="http://purl.org/rss/1.0/modules/content/"
	xmlns:dc="http://purl.org/dc/elements/1.1/"
	xmlns:atom="http://www.w3.org/2005/Atom"
	xmlns:sy="http://purl.org/rss/1.0/modules/syndication/"
		>
<channel>
	<title>Comments on: stop messing with the tablespace</title>
	<atom:link href="http://mituzas.lt/2009/05/21/innodb-tablespace/feed/" rel="self" type="application/rss+xml" />
	<link>http://mituzas.lt/2009/05/21/innodb-tablespace/</link>
	<description>where ideas come and die</description>
	<lastBuildDate>Fri, 26 Feb 2010 17:15:26 +0000</lastBuildDate>
	<sy:updatePeriod>hourly</sy:updatePeriod>
	<sy:updateFrequency>1</sy:updateFrequency>
	<generator>http://wordpress.org/?v=3.0-alpha</generator>
	<item>
		<title>By: Mark Callaghan</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189314</link>
		<dc:creator>Mark Callaghan</dc:creator>
		<pubDate>Thu, 04 Jun 2009 00:23:40 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189314</guid>
		<description>And ext-* serialize part of the read code path -- the lock is held until the file system code submits the IO request to the next layer down. sysbench fileio and SSD on xfs versus ext-2 will show the impact of that.</description>
		<content:encoded><![CDATA[<p>And ext-* serialize part of the read code path &#8212; the lock is held until the file system code submits the IO request to the next layer down. sysbench fileio and SSD on xfs versus ext-2 will show the impact of that.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Domas Mituzas</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189292</link>
		<dc:creator>Domas Mituzas</dc:creator>
		<pubDate>Sat, 30 May 2009 04:51:21 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189292</guid>
		<description>Justin, all except XFS (and even XFS does serialize if it has cached pages.. ;)</description>
		<content:encoded><![CDATA[<p>Justin, all except XFS (and even XFS does serialize if it has cached pages.. ;)</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Justin Swanhart</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189291</link>
		<dc:creator>Justin Swanhart</dc:creator>
		<pubDate>Fri, 29 May 2009 17:55:54 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189291</guid>
		<description>Which Linux filesystems serialize DirectIO writes?  I can&#039;t seem to find information about this.</description>
		<content:encoded><![CDATA[<p>Which Linux filesystems serialize DirectIO writes?  I can&#8217;t seem to find information about this.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Log Buffer #148: a Carnival of the Vanities for DBAs &#124; Pythian Group Blog</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189288</link>
		<dc:creator>Log Buffer #148: a Carnival of the Vanities for DBAs &#124; Pythian Group Blog</dc:creator>
		<pubDate>Fri, 29 May 2009 16:47:22 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189288</guid>
		<description>[...] Mituzas is having none of it. Stop messing with the tablespace, he writes. But his readers might still think [...]</description>
		<content:encoded><![CDATA[<p>[...] Mituzas is having none of it. Stop messing with the tablespace, he writes. But his readers might still think [...]</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Ronald Bradford</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189246</link>
		<dc:creator>Ronald Bradford</dc:creator>
		<pubDate>Fri, 22 May 2009 09:33:21 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189246</guid>
		<description>The art of architecture design, capacity planning and most importantly ACCURATE MONITORING of your systems physical resources and the even most important in this topic of database size and growth regularly over time is the common problem here.

There is simply too many systems where the combination of  poor design and the current limitations (or weakness) of MySQL leads to these types of issues.

We would all do better to promote the Best Practices of avoiding these types of issues in the first place.


http://ronaldbradford.com
MySQL Expert
MySQL Community Member of the Year 2009</description>
		<content:encoded><![CDATA[<p>The art of architecture design, capacity planning and most importantly ACCURATE MONITORING of your systems physical resources and the even most important in this topic of database size and growth regularly over time is the common problem here.</p>
<p>There is simply too many systems where the combination of  poor design and the current limitations (or weakness) of MySQL leads to these types of issues.</p>
<p>We would all do better to promote the Best Practices of avoiding these types of issues in the first place.</p>
<p><a href="http://ronaldbradford.com" rel="nofollow">http://ronaldbradford.com</a><br />
MySQL Expert<br />
MySQL Community Member of the Year 2009</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Peter Zaitsev</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189245</link>
		<dc:creator>Peter Zaitsev</dc:creator>
		<pubDate>Fri, 22 May 2009 00:14:04 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189245</guid>
		<description>Domas,

I read it as &quot;Perfect people never need Innodb File Per Table&quot; which is correct, however there are not much of these perfect people around. 

It is so frequent to see mistakes causing table space to run away and confuse all space on partition and  whatever dumping tools you use dumping hundreds of gigs of highly indexes data is not fun. 

Same applies to filesystem - many people just have what they have - running MySQL in hosted envinronment or having OS which does not support smart filesystems out of the box. 

Though indeed if you have many thousands of small tables file per table is quite likely to be slower.</description>
		<content:encoded><![CDATA[<p>Domas,</p>
<p>I read it as &#8220;Perfect people never need Innodb File Per Table&#8221; which is correct, however there are not much of these perfect people around. </p>
<p>It is so frequent to see mistakes causing table space to run away and confuse all space on partition and  whatever dumping tools you use dumping hundreds of gigs of highly indexes data is not fun. </p>
<p>Same applies to filesystem &#8211; many people just have what they have &#8211; running MySQL in hosted envinronment or having OS which does not support smart filesystems out of the box. </p>
<p>Though indeed if you have many thousands of small tables file per table is quite likely to be slower.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Eric Bergen</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189244</link>
		<dc:creator>Eric Bergen</dc:creator>
		<pubDate>Thu, 21 May 2009 20:53:32 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189244</guid>
		<description>The two main reasons for using file per table are...

1. alter table
2. LVM/Filesystem snapshot backups. 

LVM backups being directly related to alter table. Without using file per table if I alter a huge table now my backups have to deal with processing what is essentially wasted space. 

When online backups are better and online alter table is the norm it may make sense to ditch file per table. Until then I&#039;m sticking with it.</description>
		<content:encoded><![CDATA[<p>The two main reasons for using file per table are&#8230;</p>
<p>1. alter table<br />
2. LVM/Filesystem snapshot backups. </p>
<p>LVM backups being directly related to alter table. Without using file per table if I alter a huge table now my backups have to deal with processing what is essentially wasted space. </p>
<p>When online backups are better and online alter table is the norm it may make sense to ditch file per table. Until then I&#8217;m sticking with it.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Didier Spezia</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189241</link>
		<dc:creator>Didier Spezia</dc:creator>
		<pubDate>Thu, 21 May 2009 19:02:53 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189241</guid>
		<description>Interesting debate ... What about the number of fsync operations? If O_DIRECT is not used, having multiple files should also multiply the number of fsync (synchronous) operations ... If fsync is used instead of fdatasync, the impact on performance may be even higher.</description>
		<content:encoded><![CDATA[<p>Interesting debate &#8230; What about the number of fsync operations? If O_DIRECT is not used, having multiple files should also multiply the number of fsync (synchronous) operations &#8230; If fsync is used instead of fdatasync, the impact on performance may be even higher.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Domas Mituzas</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189240</link>
		<dc:creator>Domas Mituzas</dc:creator>
		<pubDate>Thu, 21 May 2009 17:36:44 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189240</guid>
		<description>Shlomi, 

the &#039;annoying&#039; bit wasn&#039;t personal :) It was applied to much wider group of people!

and FUD stands for &#039;fear, uncertainty, doubt&#039;. in this case, against large files! :) 

Anyway, I understand that one can come up with &#039;situations&#039;. Maybe thats unfortunate in situations with &#039;poor users&#039;. Still, if you do ALTER once, you may need to do it twice. If you have logging data, there may be more logging data, etc. 

Oh, and in my practice, I&#039;m very much used to reimaging servers. It is part of regular staging process :-)</description>
		<content:encoded><![CDATA[<p>Shlomi, </p>
<p>the &#8216;annoying&#8217; bit wasn&#8217;t personal :) It was applied to much wider group of people!</p>
<p>and FUD stands for &#8216;fear, uncertainty, doubt&#8217;. in this case, against large files! :) </p>
<p>Anyway, I understand that one can come up with &#8217;situations&#8217;. Maybe thats unfortunate in situations with &#8216;poor users&#8217;. Still, if you do ALTER once, you may need to do it twice. If you have logging data, there may be more logging data, etc. </p>
<p>Oh, and in my practice, I&#8217;m very much used to reimaging servers. It is part of regular staging process :-)</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Morgan Tocker</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189239</link>
		<dc:creator>Morgan Tocker</dc:creator>
		<pubDate>Thu, 21 May 2009 17:07:57 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189239</guid>
		<description>Related to Bill&#039;s point - I&#039;ve found it pretty common that some application logging system ends up accounting for 90% of the data.  If you delete/archive that logging data, you may do so with the intention of never reaching that same data size again.

In this case, being able to recover the space in a hurry is a nice to have ;)  It makes backups easier, etc.</description>
		<content:encoded><![CDATA[<p>Related to Bill&#8217;s point &#8211; I&#8217;ve found it pretty common that some application logging system ends up accounting for 90% of the data.  If you delete/archive that logging data, you may do so with the intention of never reaching that same data size again.</p>
<p>In this case, being able to recover the space in a hurry is a nice to have ;)  It makes backups easier, etc.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Shlomi Noach</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189237</link>
		<dc:creator>Shlomi Noach</dc:creator>
		<pubDate>Thu, 21 May 2009 16:34:02 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189237</guid>
		<description>Domaz,

1st, Three big cheers for calling me annoying, this is the first time anyone has taken my technical writings to the personal level. I don&#039;t take it hard though.

2nd: FUD? As in MS vs. Linux? FUD against who?

Now, let&#039;s describe some real scenario, and I agree with the example provided by Bill Karwin on 1st comment.
You have a database in one single tablespace. There is one very large table which takes 40% of disk space, the rest of the tables take 10%. Total 50% disk space.
Aha, now there&#039;s a need for some ALTER TABLE on the big table. ALTER TABLE requires space for building the new table. Space taken from the file system. There goes another 40% of disk space, and all of the sudden you are at 90% capacity.

Is this space I&#039;m going to utilize later on? No. I just happened to require it for a one time operation, it&#039;s not like my data will grow into it in the next couple of days.

I&#039;ve seen poor users who didn&#039;t have innodb-file-per-table, who didn&#039;t know it exists (no one told them), who got their production systems&#039;s disk full or almost full, and who had to migrate to a slave or otherwise recover from backup to regain disk space. Yes, they needed the disk space!

&quot;Oh, wait, this was given as an argument:... ...Right&quot;
English is not my mother tongue. I meant to say you don&#039;t need access for MySQL as you would need in order to access SHOW TABLE STATUS etc.</description>
		<content:encoded><![CDATA[<p>Domaz,</p>
<p>1st, Three big cheers for calling me annoying, this is the first time anyone has taken my technical writings to the personal level. I don&#8217;t take it hard though.</p>
<p>2nd: FUD? As in MS vs. Linux? FUD against who?</p>
<p>Now, let&#8217;s describe some real scenario, and I agree with the example provided by Bill Karwin on 1st comment.<br />
You have a database in one single tablespace. There is one very large table which takes 40% of disk space, the rest of the tables take 10%. Total 50% disk space.<br />
Aha, now there&#8217;s a need for some ALTER TABLE on the big table. ALTER TABLE requires space for building the new table. Space taken from the file system. There goes another 40% of disk space, and all of the sudden you are at 90% capacity.</p>
<p>Is this space I&#8217;m going to utilize later on? No. I just happened to require it for a one time operation, it&#8217;s not like my data will grow into it in the next couple of days.</p>
<p>I&#8217;ve seen poor users who didn&#8217;t have innodb-file-per-table, who didn&#8217;t know it exists (no one told them), who got their production systems&#8217;s disk full or almost full, and who had to migrate to a slave or otherwise recover from backup to regain disk space. Yes, they needed the disk space!</p>
<p>&#8220;Oh, wait, this was given as an argument:&#8230; &#8230;Right&#8221;<br />
English is not my mother tongue. I meant to say you don&#8217;t need access for MySQL as you would need in order to access SHOW TABLE STATUS etc.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Domas Mituzas</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189236</link>
		<dc:creator>Domas Mituzas</dc:creator>
		<pubDate>Thu, 21 May 2009 15:45:05 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189236</guid>
		<description>Bill, I can suggest a good book for the Manager and Programmer to read, &quot;The Art of Capacity Planning: Scaling Web Resources&quot; by John Allspaw ;-)</description>
		<content:encoded><![CDATA[<p>Bill, I can suggest a good book for the Manager and Programmer to read, &#8220;The Art of Capacity Planning: Scaling Web Resources&#8221; by John Allspaw ;-)</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Bill Karwin</title>
		<link>http://mituzas.lt/2009/05/21/innodb-tablespace/comment-page-1/#comment-189235</link>
		<dc:creator>Bill Karwin</dc:creator>
		<pubDate>Thu, 21 May 2009 15:33:28 +0000</pubDate>
		<guid isPermaLink="false">http://dammit.lt/?p=500#comment-189235</guid>
		<description>Your points are true, but here&#039;s the common scenario where file-per-table has an advantage:

Programmer: &quot;We&#039;re running out of disk space, that&#039;s why our uploads directory can&#039;t take any more files.&quot;

Manager: &quot;We have to get our website back online in the next few minutes or our business is sunk.  Archive some of the older data and delete it out of the database.&quot;

Programmer (after some minutes): &quot;Done.  But the filesystem is still full and we&#039;re still getting errors.&quot;

Manager: &quot;WTF?!?&quot;

You try telling that manager that InnoDB will reuse the space for new data gradually.  That&#039;s not going to help make space on the filesystem in the next few minutes -- or ever.</description>
		<content:encoded><![CDATA[<p>Your points are true, but here&#8217;s the common scenario where file-per-table has an advantage:</p>
<p>Programmer: &#8220;We&#8217;re running out of disk space, that&#8217;s why our uploads directory can&#8217;t take any more files.&#8221;</p>
<p>Manager: &#8220;We have to get our website back online in the next few minutes or our business is sunk.  Archive some of the older data and delete it out of the database.&#8221;</p>
<p>Programmer (after some minutes): &#8220;Done.  But the filesystem is still full and we&#8217;re still getting errors.&#8221;</p>
<p>Manager: &#8220;WTF?!?&#8221;</p>
<p>You try telling that manager that InnoDB will reuse the space for new data gradually.  That&#8217;s not going to help make space on the filesystem in the next few minutes &#8212; or ever.</p>
]]></content:encoded>
	</item>
</channel>
</rss>
