We are in the process og Oracle 11g Installation with ASM keeping OCR and VD in ASM. For external redundancy whats the best raid level for OCR and Voting Disk.
Best is kind of subjective.
My preference is for RAID10, and if that not available, RAID5. Or alternatively, no RAID at all and using normal and high redundancy via ASM. (currently using all these in a number of different RAC configs)
It is not an issue whether the raid-controller is good or bad. The issue is performance - what are the overheads and how does it affect latency? As RAID5 requires a checksum calculation on every I/O write, this is a large overhead. If that checksum calculation is done serially, then the writer needs to wait for checksum-calc-and-write to complete too. If instead the writer writes to a storage server cache (instant return of I/O write call), followed by the storage system doing the checksum-calc-and-write as part of a lazy write to disk, then the overhead does not exist from the writer's perspective.
RAID5 and RAID6 can kill database performance.
> For external redundancy whats the best raid level for OCR and Voting Disk.
If you are concerned only about store OCR and Voting Disk, from my point of view you can use any RAID (except 0) that will not make difference about performance.
This is because OCR and Voting Disk has low I/O activity.
But you must avoid place Voting Disk in a overloaded array because a Node Fencing can happens if a node not being able to ping the cluster voting files or occurs I/O starvation which makes generally ping operations not possible in a timely manner.
e.g Placing voting disk in a array where redo/datafile has high I/O activity, this can cause a problem and should be avoided.
I recommend you create a small diskgroup with Normal Redundancy to store only Voting Disk and OCR using 3 Luns preferable each Lun in different array.
1 - Separate clusterware files from database files, this prevents human errors or management issues.
e.g Cloning a Diskgroup (to Clone a database) using third-party tools, add or remove luns of diskgroup to change size or migration, drop database Files manually on ASM, etc.
2 - Multiplexing voting disk in a 3 different Luns (Normal Redundancy) you can protect clusterware files (voting disk, ocr and asmspfile) against logical/physical corruption. Using one Voting disk you are unprotected against this type of failures.
My point here limits on Clusterware World (what means which not involves Oracle RAC)
Billy has a good point, I agree with your points regarding about database perfomance and RAID.