Mirroring Strategy

[Darren Loh]

Hi experts,

I'm still new to HP-UX and HP-UX mirroring.

I have 2 questions here:

Question 1. Mirroring Strategy
Currently I have 4 hard disks (eg: disk1, disk2 , disk3 , disk4) for my Itanium server.
HP Mirror/UX software is available.

Vol group: vg00
Disk1: Being used currently
Disk2, Disk3, Disk4 are empty and new.

Disk1 has enough space for my application to run smoothly. For the remaining 3 disks, I'm seeking your expertise and advice to make full use of it.

Option 1:
Mirror the disk1 on disk2 and disk3.
Use disk4 as spare if anyone of disk1 to disk3 fail? Will multiple mirroring affect the performance?

Option 2:
Add disk2 to vg00 and extend whole hardisk space to /opt. Then, mirror disk1 on disk3 and mirror disk2 and disk4.

Other suggestions?

Question 2:
I have read When_Good_Disks_Go_Bad_WP which gives me steps on how to do mirroring but it doesn't explain much on multiple mirroring.
eg:
lvextend -m 1 /dev/vg00/lvol1 /dev/dsk/disk2
If I want to mirror the lvol1 on two disks, is the command
lvextend -m 2 /dev/vg00/lvol1 /dev/dsk/disk2 /dev/dsk/disk3

I might consider HP SMH if it works for multiple mirroring.


Please advise me.

Thank you all for your time.

BR//Darren

[James R. Ferguson]

Hi:

You don't say what kind of server you have. Newer Itanium servers come with hardware mirroring (RAID) and your vg00 would already be mirrored thusly. One way to tell this is that the disk is reported as "IR Volume" in an 'ioscan'.

That said, first, keep in mind that LVM mirroring is implemented at the logical volume level.

It is best for recovery, cloning, and (perhaps) performance) that your vg00 be composed only of the standard HP-UX operating system layout. That is, do _not_ use vg00 to hold application software and data. Not counting physical volumes used for mirroring, keeping vg00 to one physical volume rather than more than one is easier to manage.

I would certainly mirror vg00 using two physical disks. Given your description of a 4-disk system, you could then create a second volume group (vg01) for your application and mirror its logical volumes between its two disks.

Regards!

...JRF...

[VK2COT]

Hello,

a) Firstly, check if those four physical
volumes reside on the same or different
controllers.

Preferably, they should be on different controllers (at least two of them).

I really hate what most vendors do these
days: deliver servers with a single
controller.

b) So, disk1 is your primary boot disk. Use
disk2 to set up RAID-1. There are many useful
scripts to do it. Since I dislike manual
and repetitive work, I wrote a script for it:

http://www.circlingcycle.com.au/Unix-sources/HP-UX-11x-LVM-mirroring.sh.txt

I am sure you can find other references on
the net.

c) If you have no other important role for
disk3 and disk4, you can set up Dynamic Root
Disk on one or both of them for cloning.

That way you have decent RAID-1 for current boot environment, and full RAID-1 clone for
on-line patching and other work.

Simple and easy.

I would not go overboard and create two-way or three-way mirroring. Too much redundancy
for no special benefits.

Cheers,

VK2COT

[Darren Loh]

Sorry about the missing info, I don't have hardware RAID so I have to use the software Mirror/UX.

"do _not_ use vg00 to hold application software and data", I will take note of that in the future but currently all application software and data are already created and installed on the same hard disk (disk1).

[James R. Ferguson]

Hi (again):

> currently all application software and data are already created and installed on the same hard disk (disk1).

I would advise you to consider migrating your application software and data to a volume group that is separate from vg00. Mirror all the logical volumes of vg00 and mirror all the logical volumes of the new vg01.

In the event that you don't have a large enough physical volume to support all application software and application data, _at least_ try to move the volatile application data to its own volume group, leaving the static application executables on vg00. Should you ever need/want to recover vg00 from an Ignite image, you want the cleanest, most straight-forward recovery possible.

A phased implementation of a better strategy is better than no change at all.

Regards!

...JRF...

[Steven E. Protter]

Shalom Darren,

My first mirroring strategy is to use the built in smart card hardware mirror technology that many Itanium servers have.

It is superior to mirror/ux because it has its own CPU for managing I/O and if a disk fails there is an indicator on the front of the server, and hot swap replacement is easy.

This methodology is invisible to the OS. I'd set this up with two disks to boot the OS and the other two disks as needed, either raid 5 or 1, depending on application needs.

If you go with a mirror/ux strategy, I would still have vg00 for Os only, mirror the entire thing.

I would put application data on a second volume group and devote the remaining disks to data volume group. Strategy wise, I would go with single mirror copy for write intensive databases, and a raid 5 strategy for more read focused applications.

I've never felt the need in 11 years for multiple mirror copies of anything. Surely such a strategy would work, but its overkill and not a replacement for a good backup configuration.

SEP

[Darren Loh]

Dear all,

Thank you so much for the prompt yet informative advice. I will take all advice into consideration and see what I can do with current server.

Too bad, the server is already in production mode so I can't try much now.

I will be careful with my planning in the future :)

Thank you again.

Warmest regards,
Darren

[James R. Ferguson]

Hi Darren:

As a new member joining this community, please read:

http://forums11.itrc.hp.com/service/forums/helptips.do?#28

Assigning points is both a way of saying "thank you" and a way of pointing future readers to suggestions and solutions that helped solve a problem.

Regards!

...JRF...

[Steven E. Protter]

Shalom,

I urge you to check your hardware availability on the P400 smart card.

Then schedule a down time window to implement hardware raid.

It is worth it.

I echo JRF's statement. Welcome to ITRC. I've been a member since 2001 and an active participant since 2003.

SEP