Hi James, Thanks for the very informative response. James Manning wrote: > > [root] > > Hi, > > Hello. > > > I've created mirrored striped arrays (Raid10) and am not confident that > > my first striped set is in fact being mirrored on my second striped set. > > First question: did you make backups? :) This is not a production system, but a test-bed to prove production methods. My motto here is "break it early and often." I am also fairly new to Linux, so I am climbing up the curve. Gotta have those command-line skills. > > > When the mirrored mdX devices are created, cat /proc/mdstat does show > > that re-synching is taking place. However, if I mount an mdX that is > > part of my second striped set, I see NO files, just a lost+found > > directory. Hmm, I didn't mount as read-only. It this significant? > > Any chance we could see your /proc/mdstat output? I'm not on-site today, but I will send a copy ASAP. > > > What techniques can I use to verify that the second striped set is being > > mirrored? Is there a raidtool to force resynching? > > mkraid'ing md10-14 will need to write to the ends of md0-9, possibly > corrupting the filesystems already in place (with the blessed data > being on md0-4, it would appear). I created all the mdx devices at the same time. I formatted md0 - md9 as ext2, but did not run mk2fs on the mirrored devices (md10 -md 14) Next, I mounted md 0 - 4, copied over the data, changed my fstab and rebooted. At one point I "raidstop"-ed the mirrored devices and then had to "force" ;) the mkraid to recreate them. I am able to switch back and forth between the hda mount points and the mdX mount points (md0 - md4)by changing fstab and rebooting. > > Although it's not broken out as a separate section, the method for getting > a mirror made of already in-place data isn't extremely nice, but it has > been effective for many in the past. It's covered as "Method 2" at: > > http://www.linuxdoc.org/HOWTO/Software-RAID-HOWTO-4.html#ss4.12 > > If you have an ext2 resizer that you trust to shrink the fs enough for > the raid superblock, you can try that and avoid the step of copying over > data manually. Not recommended, of course, but it's a possibility. > > > If, perchance, an mdX on the first-striped set has a problem, will the > > mirrored device kick in and re-synch the striped mdX with the problem? > > When this happens (as I'm sure it probably will at some point), how will > > I know that it is occurring? I am guessing that the first striped set > > will be out of operation until it is repaired by re-synching with the > > mirrored set. > > > > How can mirroring be effectively used & monitored? > > The major problem here is that once you create (via the failed-disk method) > the raid10, you *need* to start mounting the md10-14 devices. Manually > dealing with the underlying md0-9 devices isn't supported after that point. > > It boils down to the fact that raid1 is "write to md10, mirror the > writes across md0 and md5" and not "the raid1 module should catch all > writes to md0 and automatically mirror them to md5". You have to use the > raid1 mdX device you created or you best-case lose raid1 functionality, > worst-case lose data. I guess I was under the impression that I could use the Raid0 devices as my mountpoints and have the mirrored devices take care of the 2nd stripe set as some kind of a backup or fail-over. Essentially, what you are saying is that I must mount the mirrored devices. Unfortunately, the mirrored mdX performance I'm seeing using hdparm is from 4 - 6 MBps, as opposed to 35 MBps from the striped devices. Thanks for taking the time to explain how mirroring works. FWIW, my goal is to set up a test environment of our production system. When I have confidence in our apps on the test environment I will apply the lessons learned to migrating our servers from NT to Linux. At least for the test environment, I think that using the 3ware card with their fast mirroring is looking very attractive on a cost/performance basis. > > > fstab file: > > > > /dev/md1 /local ext2 defaults 1 2 > > /dev/md0 /opt ext2 defaults 1 2 > > /dev/md4 /tmp ext2 defaults 1 2 > > /dev/md2 /usr ext2 defaults 1 2 > > /dev/md3 /var ext2 defaults 1 2 > > After the "method 2" (failed-disk) steps to get the mirrored/striped > raid10's up and running, you'll need to change these by "adding 10" to > each (md11, md10, md14, md12, md13) so you're using the raid10 devices > and not an underlying raid0 device. > > HTH, HAND > > James > -- > James Manning <[EMAIL PROTECTED]> > GPG Key fingerprint = B913 2FBD 14A9 CE18 B2B7 9C8E A0BF B026 EEBB F6E4 > > ------------------------------------------------------------------------ > Part 1.2Type: application/pgp-signature
