<div class="gmail_quote">On Thu, Mar 3, 2011 at 12:07 AM, Alex Butcher <span dir="ltr"><<a href="mailto:mythlist@assursys.co.uk">mythlist@assursys.co.uk</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im">On Wed, 2 Mar 2011, Fedor Pikus wrote:<br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
I don't think it's really an issue for RAID-1, if one disk has a glitch md<br>
will just try to update it from the other one, so you will see a resync.<br>
</blockquote>
<br></div>
md shouldn't do an entire resync of the array, just a refresh of the<br>
sector(s) that produced read errors. Bear in mind that this is new behaviour<br>
from post-2.6.15 or so.<div class="im"><br></div></blockquote><div><br></div><div>Hard to say what exactly it's doing, /proc/mdstat shows resync in progress but all drives are in the array, [UU] not [U_]. </div><div>
<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;"><div class="im">
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
md will periodically resync the array anyway (at least it does that in<br>
recent versions, my Centos5 boxes all do it), so every now and then you<br>
can catch resyncs in /proc/mdstat.<br>
</blockquote>
<br></div>
That's a distro thing, rather than md per se. There'll be a cron job<br>
(/etc/cron.weekly/99-raid-check on Fedora) that is configured with<br>
/etc/sysconfig/raid-check that does a periodic RAID scrub to find and<br>
refresh sectors that generate read errors before you, the user, do.</blockquote><div><br></div><div>Yes, that cron job is there.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im"><br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Once I had a SATA error, the disk did not drop out but md reconstructed<br>
the mirror.<br>
</blockquote>
<br></div>
Were you using a <2.6.15 kernel?<div class="im"><br></div></blockquote><div><br></div><div>2.6.18.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im">
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Since RAID-1 is a simple mirror, with each disk holding a valid<br>
filesystem, nothing will go wrong unless both disks suffer hard (not<br>
intermittent) failure in a short time. The problem with RAID-5 is that if<br>
a resync is triggered and then another disk has an intermittent failure<br>
during read, now md cannot resync the array. If you have a 2TB disk, the<br>
chances of a single read error during the read of the whole 2TB are pretty<br>
high (some estimates put it close to 100%).<br>
</blockquote>
<br></div>
Depends on the manufacturer-specified error rate. <1 in 10^14 is about<br>
11.4TiB, <1 in 10^15 is about 113.7TiB.<div class="im"><br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
I've seen recommendations to use RAID-1 or RAID-6 with large disks,<br>
instead of RAID-5 (I've done that with my main server, 6 1.5TB disks). <br>
</blockquote>
<br></div>
I don't like parity RAID, but that's because I've seen poor write<br>
performance be a botttleneck in real-world performance. Yes, capacity/£ is<br>
</blockquote><div><br></div><div>I have a RAID5 made of old 750G WD drives (which were actually some of the faster drives when I bought them), write speed is about 130-140MB/s. A pair of Seagate 1.5TB Barracudas in RAID-0 give 205MB/s. How much faster do you want in a desktop?</div>
<div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">better with parity levels, but my view is that if availability is important<br>
enough to RAID, it's important enough to use RAID1 or RAID10 (and doubly so<br>
if the choice is between a RAID5/6 made up of small and expensive SCSI<br>
drives, vs a bunch of SATA WD RE drives).<br>
<br>
Best Regards,<br><font color="#888888">
Alex</font><br>_______________________________________________<br>
mythtv-users mailing list<br>
<a href="mailto:mythtv-users@mythtv.org">mythtv-users@mythtv.org</a><br>
<a href="http://www.mythtv.org/mailman/listinfo/mythtv-users" target="_blank">http://www.mythtv.org/mailman/listinfo/mythtv-users</a><br>
<br></blockquote></div><br><br clear="all"><br>-- <br>Fedor G Pikus (<a href="mailto:fpikus@gmail.com">fpikus@gmail.com</a>)<br><a href="http://www.pikus.net">http://www.pikus.net</a><br><a href="http://wild-light.com">http://wild-light.com</a><br>