<?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: Taste test:  Innobackup vs. Xtrabackup</title>
	<atom:link href="http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/feed/" rel="self" type="application/rss+xml" />
	<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/</link>
	<description>News and views from Pythian DBAs</description>
	<lastBuildDate>Thu, 18 Mar 2010 23:57:39 +0000</lastBuildDate>
	<generator>http://wordpress.org/?v=2.8.4</generator>
	<sy:updatePeriod>hourly</sy:updatePeriod>
	<sy:updateFrequency>1</sy:updateFrequency>
		<item>
		<title>By: Massimo</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-397721</link>
		<dc:creator>Massimo</dc:creator>
		<pubDate>Tue, 12 Jan 2010 09:03:43 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-397721</guid>
		<description>Hi at all,
someone know it&#039;s possible to backup stored procedure with innobackupex when I use --database=only_one_db</description>
		<content:encoded><![CDATA[<p>Hi at all,<br />
someone know it&#8217;s possible to backup stored procedure with innobackupex when I use &#8211;database=only_one_db</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Sheeri Cabral</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-380854</link>
		<dc:creator>Sheeri Cabral</dc:creator>
		<pubDate>Mon, 26 Oct 2009 22:06:41 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-380854</guid>
		<description>si -- you&#039;ll have to file a bug with xtrabackup for that, I can&#039;t confirm that and I don&#039;t want people to think this is true when it may be only on your system, or the bug may already be fixed.</description>
		<content:encoded><![CDATA[<p>si &#8212; you&#8217;ll have to file a bug with xtrabackup for that, I can&#8217;t confirm that and I don&#8217;t want people to think this is true when it may be only on your system, or the bug may already be fixed.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: si</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377823</link>
		<dc:creator>si</dc:creator>
		<pubDate>Fri, 02 Oct 2009 22:02:47 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377823</guid>
		<description>xtrabackup not working with innodb plugin. with xtrabackup + innodb plugin you can lost you data.</description>
		<content:encoded><![CDATA[<p>xtrabackup not working with innodb plugin. with xtrabackup + innodb plugin you can lost you data.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Sheeri Cabral</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377790</link>
		<dc:creator>Sheeri Cabral</dc:creator>
		<pubDate>Fri, 02 Oct 2009 15:51:24 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377790</guid>
		<description>Mark, thanx for that.  We compared iostat and swap activity and found that xtrabackup was (slightly) more resource intensive for a (slightly) shorter period of time than InnoDB Hot Backup.

For example, there were about 2,000 more transfers per second using xtrabackup, and the heavy I/O dropped off 20 minutes sooner.

With swapping, sar reported pswapin/sec as slightly higher for xtrabackup (0.04 vs 0.03) and the swapping dropped off 15 minutes sooner.

The query response time was actually worse with xtrabackup, but only slightly so.  This is likely due to xtrabackup using more resources than InnoDB Hot backup -- since it only uses slightly more resources, the query response time is only slightly worse.  This is a heavily used server, and the degradation isn&#039;t affecting the user experience (partly because the hot backups are done during a low activity time).</description>
		<content:encoded><![CDATA[<p>Mark, thanx for that.  We compared iostat and swap activity and found that xtrabackup was (slightly) more resource intensive for a (slightly) shorter period of time than InnoDB Hot Backup.</p>
<p>For example, there were about 2,000 more transfers per second using xtrabackup, and the heavy I/O dropped off 20 minutes sooner.</p>
<p>With swapping, sar reported pswapin/sec as slightly higher for xtrabackup (0.04 vs 0.03) and the swapping dropped off 15 minutes sooner.</p>
<p>The query response time was actually worse with xtrabackup, but only slightly so.  This is likely due to xtrabackup using more resources than InnoDB Hot backup &#8212; since it only uses slightly more resources, the query response time is only slightly worse.  This is a heavily used server, and the degradation isn&#8217;t affecting the user experience (partly because the hot backups are done during a low activity time).</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Mark Callaghan</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377785</link>
		<dc:creator>Mark Callaghan</dc:creator>
		<pubDate>Fri, 02 Oct 2009 15:18:01 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377785</guid>
		<description>I am not a Linux kernel hacker nor an expert on the Linux VM, but I can share a bit more background from my peers who submitted the patch to use POSIX fadvise. vmstat showed too much activity for the si and/or so columns when hot backups were done. The theory is that recently read data from the files copied during hot backup pushed other pages out of physical memory and written to the swap space on disk. The fear is that this included pages from the InnoDB buffer cache. So even though O_DIRECT was used, scanning large files had an impact on InnoDB performance. When xtrabackup was changed to use POSIX fadvise, swap activity decreased as listed in vmstat and query response time improved.</description>
		<content:encoded><![CDATA[<p>I am not a Linux kernel hacker nor an expert on the Linux VM, but I can share a bit more background from my peers who submitted the patch to use POSIX fadvise. vmstat showed too much activity for the si and/or so columns when hot backups were done. The theory is that recently read data from the files copied during hot backup pushed other pages out of physical memory and written to the swap space on disk. The fear is that this included pages from the InnoDB buffer cache. So even though O_DIRECT was used, scanning large files had an impact on InnoDB performance. When xtrabackup was changed to use POSIX fadvise, swap activity decreased as listed in vmstat and query response time improved.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Mark Callaghan</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377782</link>
		<dc:creator>Mark Callaghan</dc:creator>
		<pubDate>Fri, 02 Oct 2009 15:04:40 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377782</guid>
		<description>The use of POSIX fadvise should benefit MyISAM queries, not the backup time. So the difference would be measured in query response time and iostat results measured during the backup window.</description>
		<content:encoded><![CDATA[<p>The use of POSIX fadvise should benefit MyISAM queries, not the backup time. So the difference would be measured in query response time and iostat results measured during the backup window.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Sheeri Cabral</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377781</link>
		<dc:creator>Sheeri Cabral</dc:creator>
		<pubDate>Fri, 02 Oct 2009 14:53:22 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377781</guid>
		<description>We didn&#039;t see any impact other than the 1% or so reported in the blog post.  We are using xtrabackup Ver 0.8.1rc Rev 78, so it seems like the lack of OS cache flushing isn&#039;t making a difference (in our case). 

mysql&gt; select count(*), sum(data_length+index_length)/1024/1024/1024 as sizeGb, engine from tables where table_schema not in (&#039;mysql&#039;,&#039;information_schema&#039;) group by engine;
+----------+------------------+--------+
&#124; count(*) &#124; sizeGb           &#124; engine &#124;
+----------+------------------+--------+
&#124;        1 &#124;             NULL &#124; NULL   &#124; 
&#124;      555 &#124; 302.903549194336 &#124; InnoDB &#124; 
&#124;       86 &#124;  20.829088966362 &#124; MyISAM &#124; 
+----------+------------------+--------+
3 rows in set (23.77 sec)

We&#039;re using plenty of MyISAM tables and InnoDB tables (the NULL table is a view).</description>
		<content:encoded><![CDATA[<p>We didn&#8217;t see any impact other than the 1% or so reported in the blog post.  We are using xtrabackup Ver 0.8.1rc Rev 78, so it seems like the lack of OS cache flushing isn&#8217;t making a difference (in our case). </p>
<p>mysql> select count(*), sum(data_length+index_length)/1024/1024/1024 as sizeGb, engine from tables where table_schema not in (&#8217;mysql&#8217;,'information_schema&#8217;) group by engine;<br />
+&#8212;&#8212;&#8212;-+&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;+&#8212;&#8212;&#8211;+<br />
| count(*) | sizeGb           | engine |<br />
+&#8212;&#8212;&#8212;-+&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;+&#8212;&#8212;&#8211;+<br />
|        1 |             NULL | NULL   |<br />
|      555 | 302.903549194336 | InnoDB |<br />
|       86 |  20.829088966362 | MyISAM |<br />
+&#8212;&#8212;&#8212;-+&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;+&#8212;&#8212;&#8211;+<br />
3 rows in set (23.77 sec)</p>
<p>We&#8217;re using plenty of MyISAM tables and InnoDB tables (the NULL table is a view).</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Sheeri Cabral</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377777</link>
		<dc:creator>Sheeri Cabral</dc:creator>
		<pubDate>Fri, 02 Oct 2009 14:32:47 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377777</guid>
		<description>Mark -- I don&#039;t know about how many versions the binary will work on, but I guess it depends on what changes between the versions -- it&#039;s really an OS thing.

I do know that I&#039;ve been able to get binaries for different operating systems when companies with perpetual licenses migrate their systems (for free).</description>
		<content:encoded><![CDATA[<p>Mark &#8212; I don&#8217;t know about how many versions the binary will work on, but I guess it depends on what changes between the versions &#8212; it&#8217;s really an OS thing.</p>
<p>I do know that I&#8217;ve been able to get binaries for different operating systems when companies with perpetual licenses migrate their systems (for free).</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: Mark Callaghan</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377774</link>
		<dc:creator>Mark Callaghan</dc:creator>
		<pubDate>Fri, 02 Oct 2009 14:07:32 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377774</guid>
		<description>The previous comment refers to the use of POSIX fadvise calls by Xtrabackup to avoid wiping the OS buffer cache when copying files as described here -- http://www.mysqlperformanceblog.com/2009/05/21/xtrabackup-07-rc. If you also use MyISAM or run things other than MySQL on the same host, this can be important.

The perpetual license for ibbackup -- http://www.innodb.com/products/hot-backup/order/ -- means you can use the binary that you have forever and get upgrades for 3 years. The current version of ibbackup doesn&#039;t support the Barracuda file format that is new in the InnoDB 5.1 plugin. Customers who want to do hot backup when using compressed InnoDB tables with Barracuda will have to purchase new licenses for ibbackup if they are beyond the 3 year upgrade limit when support for Barracuda is released in ibbackup.

Does the binary ABI for RHEL/CentOS allow a static binary to work on both RHEL/CentOS versions 4 and 5? I am curious about how long a binary can be used as OS versions are upgraded. I know Solaris is great about that.</description>
		<content:encoded><![CDATA[<p>The previous comment refers to the use of POSIX fadvise calls by Xtrabackup to avoid wiping the OS buffer cache when copying files as described here &#8212; <a href="http://www.mysqlperformanceblog.com/2009/05/21/xtrabackup-07-rc" rel="nofollow">http://www.mysqlperformanceblog.com/2009/05/21/xtrabackup-07-rc</a>. If you also use MyISAM or run things other than MySQL on the same host, this can be important.</p>
<p>The perpetual license for ibbackup &#8212; <a href="http://www.innodb.com/products/hot-backup/order/" rel="nofollow">http://www.innodb.com/products/hot-backup/order/</a> &#8212; means you can use the binary that you have forever and get upgrades for 3 years. The current version of ibbackup doesn&#8217;t support the Barracuda file format that is new in the InnoDB 5.1 plugin. Customers who want to do hot backup when using compressed InnoDB tables with Barracuda will have to purchase new licenses for ibbackup if they are beyond the 3 year upgrade limit when support for Barracuda is released in ibbackup.</p>
<p>Does the binary ABI for RHEL/CentOS allow a static binary to work on both RHEL/CentOS versions 4 and 5? I am curious about how long a binary can be used as OS versions are upgraded. I know Solaris is great about that.</p>
]]></content:encoded>
	</item>
	<item>
		<title>By: mjiang</title>
		<link>http://www.pythian.com/news/4177/taste-test-innobackup-vs-xtrabackup/#comment-377748</link>
		<dc:creator>mjiang</dc:creator>
		<pubDate>Fri, 02 Oct 2009 08:11:21 +0000</pubDate>
		<guid isPermaLink="false">http://www.pythian.com/news/?p=4177#comment-377748</guid>
		<description>If you copy big files off the system(Linux) on production server, then even with swappness=0 it is still going to flush the cache. xtrabackup has the fix, but I do not think innodb hotbackup has it.</description>
		<content:encoded><![CDATA[<p>If you copy big files off the system(Linux) on production server, then even with swappness=0 it is still going to flush the cache. xtrabackup has the fix, but I do not think innodb hotbackup has it.</p>
]]></content:encoded>
	</item>
</channel>
</rss>
