How mirror UX works.

[Juan Gonzalez_2]

We have four 73GB disk drives that belong to vg00 and vg00 has /var, /tmp, /var/adm/crash, /stand, /usr, /tmp, and /opt. We have all these directories lvols are mirror to a second disk. There is 9444 extends free that are distributed on all four disk that we can use for temporary space. There is a guy here that saids that with mirror UX those disk are physically mirrored so we can only use all of the 9444 extends that are free. He saids, " We can only use 4722. Which I think he is wrong because with Mirror UX you use the lvols to make it look like the disk are physically mirrored. Can some one verify my way of thinking Mirror UX works? See attachment of the volume group.

[Bill Hassell · Accepted solution]

The volume grouip is just a collection of extents which may be all one disk or as in your case, four disks. Mirroring indeed requires twice (or 3 times) as many extents. Normally, you woulde mirror an lvol by using extents from another disk so a failure will not cause loss of data. You can have data on 4722 extents and the other 4722 extents are realtime copies of those extents. The report from vgdisplay shows all extents since it doesn;t know what you plan to do with the free extents. If you mirror all the lvols, then you have 2x73Gb of space with 2x73Gb of mirrored data.

[Helen French]

First of all, unless I really want to, I won't put 4 73GB disks in vg00. If I want mirroring for root disks, I would put 2 73GB in vg00 (or if I have a less capacity disk like 36GB or 18GB, I would use that for vg00) and the rest of the disks in vg01 or vg02 etc. You really don't want 73GB disk for vg00 (which normally contains only root logical volumes). If you have a less capacity disk for root, it's easy during a Ignite recovery tape creation and an actual recovery and also easy file system administration.

Now, in your case what the 'other guy' said might be right. Like you said, if you have 9444 extents available and if it's distributed among all 4 disks, during a mirroring (of LV) you will only get 50% and that is 4722. The other 4722 is your mirror. Now if you really don't want to mirror your new LVs (or LV extend), you can make use of all these space 9444. But no data redundancy.

[Norman_21]

Some good Tips from the knowledgeBase!
http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000063199936

Another document from hp.
http://www.docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B7961-90026/B7961-90026_top.html&con=/hpux/onlinedocs/B7961-90026/00/00/40-con.html&toc=/hpux/onlinedocs/B7961-90026/00/00/40-toc.html&searchterms=mirror%7cdisk&queryid=20031122-002705

Good luck.

[Bonny Jose_2]

Only the Physical Extends used by the mirrored LVs are physically mirrored. Remaining PEs are not mirrored currently, and so you can make use of entire 9444 Extends. So the available space on the VG00 would be 9444*PE size(may be 4mb*9444) and can be used fully without redundancy or 4722*PE size with one mirror copy.


rgds
Bonny Jose

[Tim D Fulford]

Hi

I did not understand the positions you & your collegue have, so I've broken out some individual statemets.

> There is 9444 extends free that are
> distributed on all four disk that we can
> use for temporary space
This is exactly as you put it 9444 extents free, e.g. unused & unmirrored.

> There is a guy here that saids that with
> mirror UX those disk are physically
> mirrored so we can only use all of the
> 9444 extends that are free
This is not so the disks are NOT physically mirrired, the logical volumes are mirrored across 2 (4, 6, 8 etc) Physical volumes (disks). Thus if you want to mirror a LV of 1 extent you will require 1 extent from one physical volume (disk) & 1 extent from another. so 2 will be used in total.

I get confused here because this statement contradictes the one above!!

> He saids, "We can only use 4722".
you can only get 4722 extents worth of MIRRORED space (e.g. using 9444 extents). OR 9444 extents of UN-MIRRORED space

> Which I think he is wrong because with
> Mirror UX you use the lvols to make it
> look like the disk are physically
> mirrored.
Yes this is right, the logical volumes are mirrored onto other physical volumes (disks) & you can do it in ANY order, say Physical Volume1 has lvol1, lvol2, lvol3 on it & the lvols are mirrored onto Physical Volume2 which has lvol3, lvol2 & lvol1. **

so to re-cap.. if you have 9444 extents in the volume group you can use them to make 4722 extents worth of MIRRORED logical volumes. If you wish (which I usually do) you can order the mirrored logical volumes on the disks such that it LOOKS like the disks are mirrored (but they do not need to be).

** caveat ** (there is always one). If you want to make a bootable mirror copy of for vg00 lvol1, lvol2 & lvol3 MUST be mirrored in that order on both (all) physical volumes (disks).

Regards

Tim

[Tony Horton]

Hi You are both right to an extent. you can do as you say and use all the extents for a non mirrored lv of size 9444 extents, however if you do this the root disk won't necessarily be strictly mirrored any more (depending on where the free extents are).

If you have two disks where all of the lvols are strictly mirrored it is a much simpler afair to replace the disk if it fails, if there are non-mirrored lvols on your mirrored disks things get pretty messy when things go wrong, it is therefore advisable to fully mirror all lvols on your mirrored disks (which is what I think your coleague is getting at).

Regards,

Tony.

[Juan Gonzalez_2]

Tony Horton if one of the disk that has mirrored lvols crashes and there is part of an lvol that is not mirrored on it and does not belong to the root system, will the server crash? I really do not care if the data in this certain lvol that spans both mirrored disk is lost and mostly will be lost.

[A. Clay Stephenson]

If by "root system" you mean that a crashed lvol is not /, /var, /usr, /tmp, OR a swap device then the system should tolerate a failed LVOL without a crash. Of course, any applications that depend upon the down LVOL are toast. It's generally a far better approach to use two disks for vg00 and mirror everything and put data in another VG. This makes for much cleaner system upgrades.

[Geoff Wild]

Here's a good presentation on the MirrorDisk/UX and Online JFS.

Rgds...Geoff

[Tony Horton]

As Clay said, the system should not crash, but obviously the non mirrored filesystem would not be available.

The problem is it becomes messy when you go to replace the failed disk. If the disks are hot plug and are fully mirrored then it is possible to replace the disk without powering down the server provided the correct approach is taken. As far as I can remember this is not possible unless all of the lvols on the disk are mirrored. I think the other thing is if you have hot spares, then they can only be used if the disks that they are spares for are fully mirrored.

Note that replacing a failed hot plug disk without a restart is not for the feint hearted. The internal hot plug disks on the L class (not sure about other servers) are what HP call hot plug, but not hot swap, which means you can't just pull the failed one out and plug in the new one.

Regards,

Tony.