Triple Mirrored Set Creation

Gritz,
You cannot create a triple mirror on a CLARiiON array. Your choices (depending on model and FLARE revision) are RAID 1, 10, 3, 5, and 6.

Gritz, welcome to the forums ! Make sure you tell us something about yourself in the Coffee Break.

The only tripple mirror on a Clariion is in the "hidden" part of the first 5 drives. For the rest you cannot create a tripple or quadruple mirror on Clariion yourself. You can however create a mirrorview set up where the source replicates to 2 or 3 targets. But reading your question, that's not what you mean.

If you are looking at 3rd mirror from recovery point of view, then snapclone can be an option. Just keep it in synchronised mode and do not split.

But, still not sure if this is what you are looking for ?

My dilemma comes from the manufacturer's (IBM) terminology versus EMC's terminology. I understand a triple mirror. Is is possible to duplicate this with Snapclone? Mirrorview? The idea, (if I understand IBM correctly!) is to have the database striped across the three sets of disks (mirrored set of striped subsets). Bottom line, the database can be quiesced and a backup taken from the 3rd set of disks.
And the piece I really do not understand, if the EMC OS is using a triple mirror, why would you not be able to create one on your own disks?
In addition, there are simple, inexpensive(?) SATA NAS boxes that are 0+1 capable. Why would the more sophisticated Clariion not have the same capability? Again, there is something very obvious I am missing.

Gritz,
A triple mirror and RAID 0+1 are not the same thing. The CLARiiON supports RAID 0,1,3,5,6 and 1+0.

What is it you are trying to accomplish? If i know that I can answer your question better.

This is a synopsis from the IBM guide to disk subsystems for Universe. Unfortunately, I cannot paste a picture, but the diagram shows 3 sets of disks, striped and mirrored.
IBM refers to this as Raid 0+1. At the bottom (in italic) are the steps we need to accomplish.

© 2006-2008 IBM Corporation IBM U2 Lab Services 21
Alternative 1 ¿A Valid Backup (Step 1)
1.Quiesce the Data BaseEnsures no broken files¿SUSPEND.FILES ON is the command¿Pauses all writes¿Data can still be read¿Looks like sessions hung to users¿No loss of data being entered by users ¿it is retained in the i/o buffers
To quiesce the data server should
only takes about 10-30 seconds.
If it takes longer than that, you don¿t have enough disk drives under your system
This is the recommended minimum for all UV ShopsYou do not need to force logoff everyone to get a valid backup!
People performing reads continue until they perform their first write.
Disk Subsystem ¿Striped and Triple Mirrored
The total number of disk drives under the Primary data set will be determined after further analysis at your site.Two Separate Mirrored Disk Drives will be required for the Transaction Logs. Transaction Logs should not be placed on the same disk drives as the Primary data set.
The functional equivalent of breaking off a mirrored set in the SAN environment are:+ BCV+ Flash Copy.

1.Quiesce the Data BaseEnsures no broken files
¿Pauses all writes¿Data can still be read
¿Looks like sessions hung to users
¿No loss of data
2.Split off a set of the mirrored drives
3.Restart the Data Base
4.Backup the system to tape
5.Rejoin the mirrored drives

Alright, so the steps you require can be accomplished on a CLARiiON with an array based replication software called Snapview. It is a feature of the FLARE operating code that has to be enabled via a separately purchased license (unless it was licensed when you purchased the array).

Snapview can create either a full clone of the source volume or a snapshot. The clone/snap can be mounted to a backup server to perform the backup of the source volume.

There are a few different ways this can be done with Snapview. I would start first though by reviewing the product first. You can find whitepapers on Snapview on the Powerlink website here:
Support > Technical Documentation and Advisories > Software ~ S ~ Documentation > SnapView > White Papers

Gritz,



You don't want a RAID 0+1 ! When speaking of striped mirrors, it's RAID 10 (1+0). The performance is the same, but a RAID 0+1 is more vulnerable since you're talking about 2 stripes which are mirrored. When 1 disk fails, a whole stripe fails and all your data is left on 1 stripe. In a RAID 10 set up it's a bunch of mirrors which are striped. If a disk fails, you only loose the redundancy within one of the many mirrors.

For the rest I agree that perhaps you'd make a little bit more clear what it is that you want to accomplish.

When you have a protected LUN (RIAD5 or RAID10) which has a clone running next to it, you can stop the sync to the clone at a certain time and make a backup from the clone (or run any application against the clone) without impact on the original LUN. Whenever you want to sync the clone again to get it to have the same data as your LUN, sync it and split it again later.

AFAIK RAID 0+1 is exactly as fast as RAID 1+0.... I've never hear anyone request specifically 0+1 before especially since 0+1 is unsafer than 1+0.

Actually, RAID 0+1 is exactly what I want, or I should say, what I need for this particular database. This is for speed for database transactions. The transaction log is on separate drives. The IBM Universe database thrives on many small fast disks versus larger disks. I believe the third mirror is the key from the redundancy and backup standpoint. I understand the inherent problems with 0+1 versus 1+0, but this is the manufacturers recommendation.

Gritz,
Since the EMC arrays do not support RAID 0+1, you will have to go with RAID 1+0. The IBM requirements are probably in the context of the features of the DS series arrays and don't sound like general recommendations for storage array architectures to support the database.

I agree with RRR, I beleive the performance characteristics of RAID 0+1 and RAID 1+0 are similar, and with sufficient number of spindles behind the RAID set you should be able to meet your performance needs.

Again, a triple mirror is not a feature of RAID 0+1. If there is a vendor offering this feature, it is a unique feature and not a standard implementation of RAID. That being said, that would be a nice feature to have in a CLARiiON array .

But on the CLARiiONs you are limited to using a synchronous clone for your backup/DR requirements.

Or trade in your CX for a DMX !

In a DMX you can have tripple and even quadruple mirrors. Nobody uses them and especially not for splitting off 1 mirror anyways

For what it's needed here, I'd say a clone running beside the original data is what he's looking for.