Need advice disk fault tolerance and moving pv's

[Doug_3]

Hello,

I need some sys admin advice on an N9000. It has two internal seagates and a 12H array. I want to maximize availability and have some problems. The original installation was as follows:
vg00 (seagate1) has all mounted directories
vg00 (seagate2) has raw devices (root and logs)
vg01-02 (array) has raw devices (all other dbases)

Obviously if either of the seagates have problems the system is down until the disk is replaced and a restore occurs. Not what I want to go through.

I'd like to move everything off vg00 except / stand etc sbin and then add a third tempdbspace. Move everything off vg01 and mirror vg00 to it and add a second tempdbspace.

What are your recommendations? Plus, I assume I can use pvchange to address both mounted and raw files w/o too many complications?

Thanks,
Doug

[Murray Jones · Accepted solution]

Doug,

I'll think you'll find the normal approach is to seperate your system and app resources.

vg00 will normally be reserved solely for root disks and system functionality. In the case of your N-class: root filesystems / swap / dump live on one of the internal disks, mirrored to the other internal disk using software mirroring (mirrorUX). Don't extend vg00 outside the internal disks unless you need to for system reasons - not because you want to put app stuff on vg00 filesystems.

Keep all of your application stuff off of the internal disks and out of vg00. Don't clutter the root disks with application data - especially things like temp tablespaces that may be I/O intensive. ie, try not to mix system and app resources. Create vg01, vg02 etc in your array with like devices.

Also remember for things like mirroring, you want to mirror like for like - so mirroring your internal disks to devices outside in an array may be difficult and/or non-efficient.

By logically seperating your volume and device groups, you can minimise problems caused when a device fails. For example, you might lose a disk that has a temp tablespace on it... so you might lose the application for a while andhave to recreate that disk - but if it is in vg00, you might lose the application and the box until it is repaired.

So, how to go about it.... To do any sort of migration, you're going to need alot of free (read unallocated) space. Unless you have it, I'd suggest that perhaps a backup / destroy / reconfigure / restore option might be better suited. Either way, extensive planning will need to occur. I'd suggest that even with sufficient space for the migration method, the rebuild option may end up easier. By the sounds of your vg00 config, depending on whether your lv's extend across onto the second disk or not (ie, are they discreetly on seperate devices?),you might be able to be migrate all system related filesystems to reside on 1 physical internal disk - then vgreduce vg00 to contain only that disk, then mirror to the other internal disk.

Unfortunately, that's about as far as the device can go without getting into detailed planning.. and knowledge of your filesystem layout.

hope it helps,

Murray

[Doug_3]

thank you.