<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><br><br><div>Thank you for your explanation. I have found this in change log, Can you explain it to me?</div><div id="divNeteaseMailCard"></div>In case our backing devices support write barriers and cache flushes, we use these means to ensure data integrity in the presence of volatile disk write caches and power outages.<br><pre><br>在&nbsp;2012-06-04&nbsp;15:48:36，"Felix&nbsp;Frank"&nbsp;&lt;ff@mpexnet.de&gt;&nbsp;写道：

&gt;Hi,

&gt;

&gt;please&nbsp;don't&nbsp;forget&nbsp;to&nbsp;CC&nbsp;the&nbsp;list.

&gt;

&gt;On&nbsp;06/04/2012&nbsp;02:58&nbsp;AM,&nbsp;陈楠&nbsp;wrote:

&gt;&gt;&nbsp;Thank&nbsp;you!&nbsp;There&nbsp;are&nbsp;volatile&nbsp;write&nbsp;cashes&nbsp;in&nbsp;our&nbsp;I/O&nbsp;system.&nbsp;Does&nbsp;the

&gt;

&gt;

&gt;As&nbsp;long&nbsp;as&nbsp;you're&nbsp;making&nbsp;use&nbsp;of&nbsp;these&nbsp;caches,&nbsp;you&nbsp;*will*&nbsp;loose&nbsp;data&nbsp;on

&gt;power&nbsp;outage&nbsp;or&nbsp;critical&nbsp;hardware&nbsp;failure.

&gt;Performance&nbsp;can&nbsp;become&nbsp;unacceptable&nbsp;when&nbsp;you&nbsp;disable&nbsp;them.&nbsp;You&nbsp;can

&gt;remedy&nbsp;the&nbsp;situation&nbsp;by&nbsp;protective&nbsp;means&nbsp;such&nbsp;as&nbsp;battery&nbsp;backups&nbsp;or

&gt;(preferable&nbsp;imho)&nbsp;flash&nbsp;memory&nbsp;units&nbsp;on&nbsp;your&nbsp;RAID&nbsp;controllers.

&gt;

&gt;&gt;&nbsp;new&nbsp;version&nbsp;has&nbsp;built-in&nbsp;data&nbsp;protection?&nbsp;We&nbsp;have&nbsp;four&nbsp;pairs&nbsp;redundancy

&gt;

&gt;As&nbsp;I&nbsp;said&nbsp;-&nbsp;there&nbsp;is&nbsp;no&nbsp;such&nbsp;thing.&nbsp;DRBD&nbsp;does&nbsp;not&nbsp;know&nbsp;(or&nbsp;care)&nbsp;which

&gt;files,&nbsp;buffers&nbsp;or&nbsp;whatever&nbsp;your&nbsp;application&nbsp;is&nbsp;writing&nbsp;to.&nbsp;All&nbsp;it&nbsp;sees

&gt;is&nbsp;a&nbsp;stream&nbsp;of&nbsp;blocks.&nbsp;If&nbsp;this&nbsp;stream&nbsp;gets&nbsp;torn,&nbsp;you&nbsp;end&nbsp;up&nbsp;with&nbsp;some

&gt;data&nbsp;on&nbsp;your&nbsp;secondary,&nbsp;but&nbsp;it's&nbsp;impossible&nbsp;for&nbsp;DRBD&nbsp;to&nbsp;(e.g.)&nbsp;roll&nbsp;back

&gt;to&nbsp;the&nbsp;last&nbsp;sane&nbsp;filesystem&nbsp;state.

&gt;

&gt;&gt;&nbsp;servers,&nbsp;All&nbsp;of&nbsp;them&nbsp;powered&nbsp;off&nbsp;by&nbsp;accident,&nbsp;only&nbsp;one&nbsp;pair&nbsp;servers

&gt;&gt;&nbsp;works&nbsp;fine,&nbsp;others&nbsp;all&nbsp;crashed.&nbsp;I&nbsp;think&nbsp;there&nbsp;are&nbsp;something&nbsp;wrong&nbsp;with

&gt;&gt;&nbsp;our&nbsp;applications.&nbsp;I&nbsp;want&nbsp;to&nbsp;know&nbsp;how&nbsp;to&nbsp;adapt&nbsp;drbd&nbsp;mechanism&nbsp;for&nbsp;our

&gt;&gt;&nbsp;application.

&gt;

&gt;Describe&nbsp;your&nbsp;application&nbsp;then.&nbsp;But&nbsp;DRBD's&nbsp;MO&nbsp;cannot&nbsp;be&nbsp;altered&nbsp;much.

&gt;

&gt;Usually,&nbsp;the&nbsp;type&nbsp;of&nbsp;application&nbsp;that&nbsp;DRBD&nbsp;lends&nbsp;itself&nbsp;well&nbsp;to&nbsp;is&nbsp;one

&gt;that&nbsp;can&nbsp;recover&nbsp;from&nbsp;crashes&nbsp;by&nbsp;some&nbsp;means&nbsp;of&nbsp;journal&nbsp;or&nbsp;log.&nbsp;A

&gt;prominent&nbsp;example&nbsp;is&nbsp;the&nbsp;InnoDB&nbsp;storage&nbsp;engine&nbsp;in&nbsp;MySQL.&nbsp;A&nbsp;MyISAM

&gt;database&nbsp;will&nbsp;not&nbsp;benefit&nbsp;as&nbsp;much&nbsp;from&nbsp;DRBD,&nbsp;because&nbsp;it&nbsp;will&nbsp;likely&nbsp;be

&gt;thoroughly&nbsp;corrupted&nbsp;if&nbsp;the&nbsp;primary&nbsp;does&nbsp;crash.

&gt;

&gt;HTH,

&gt;Felix

</pre></div><br><br><span title="neteasefooter"><span id="netease_mail_footer"></span></span>