<br><br><div class="gmail_quote">On Fri, Oct 24, 2008 at 10:11 AM, Yan Seiner <span dir="ltr"><<a href="mailto:yan@seiner.com">yan@seiner.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
If you do that, could you please post your results on line somewhere? I'm<br>
in the process of getting all the pieces for a 20 disk raid box, and I<br>
will need to make some decisions in the next couple of weeks on how to<br>
structure the filesystem. Right now it looks like 4 x 5 - 1TB disk RAID5<br>
at full buildout. I'm starting with 3 1TB drives and will grow those to<br>
the first 5 disk RADI5 eventually. But if ZFS can be made to grow from 3<br>
to 5 disks, and then then stick those 5 disk bundles together, that would<br>
be really cool.</blockquote></div><br><br>I will, but I'm still a few months out most likely. I don't think you would be able to add disks to the individual raidz, you would create a new one in the pool. So you will have a second parity disk, but that's a good thing with a 20 disk RAID. I'd probably put 5 disks in each raidz and put the 4 raidz devices into a single pool. Then create the filesystems on top of all that. Then you can lose any 1 disk in each 5-disk raidz before losing data. <br>
<br>This is all based on reading the docs, I haven't tried it myself on real hardware yet. If you have zfs-fuse installed, you can use files as the block devices in ZFS (loopback filesystem images). So you can do this testing yourself on your intended platform. I might try it in VMWare myself now that I think about it. :) <br>