I have MFSYS25v2 systems in the field with 4 disks in an OS drive pool. The virtual drives for the OSs are set as RAID 10. I recently had a multiple disk failure on one of my systems that brought down the entire OS drive pool. Drives 1 and 3 failed. Normally in Raid 10 the RAID is written like this:
In RAID 10 data is striped across 1 and 3, 1 is mirrored to 2 and 3 is mirrored to 4. If drives 1 and 3 fail drives 2 and 4 have all of the required data to keep the array running.
This wasn't the case in my situation. As soon as drive 1&3 failed, all virtual drives were destroyed. It appears that RAID 10 on the Modular Server RAID controller is actually RAID 0+1. RAID 0+1 looks like this:
Notice that data is striped from Drive 1 to Drive 2. Drive 1 is mirrored to drive 3 and Drive 2 is mirrored to Drive 4. In my situation since Drive 1 and 3 failed all of my virtual drives failed. In RAID 0+1 you lose everything if drive 1&3 or drive 2&4 fail. This is not the case in RAID 10.
I am concerned that since RAID 0+1 is in use instead of RAID 10 the fault tolerance is negatively affected.
In this table it shows the RAID 10 can remain online in twice the number of scenarios as RAID 0+1. I would like clarification on this as it pertains to the Modular Server RAID controller.
The Modular Server does use RAID 0+1 for RAID 10. I think the help file even says:
RAID level 10
In RAID level 10 (or 0+1), the RAID levels are "nested."
Thanks for the reply. According to the RAID table I posted it looks like RAID 0+1 is a lot more vulnerable than RAID 10. However, I have had drives 2&3 fail in a 4 disk RAID 10 on a modular server and the Array was OK. I would expect that 1&4 could possibly fail as well. It would be nice if I could get something official from Intel that would identify which two drives could fail without the Array going down.
On RAID 0+1, you can lose drives 1 and 4, or 2 and 3.
On RAID 1+0, you can lose drives 1 and 2, or 3 and 4.
Is that your understanding as well?
I have had drives 2&3 fail on a Modular Server RAID 10 (technically 0+1) with 4 drives and the array did not go down.
So I believe the following to be true:
RAID 0+1 (What the modular server calls RAID10)
You can lose...
I am unsure of 1&4 but would lean towards expecting it to stay online as you have the Original Disk 2 and the Mirror of Disk 1 albeit they are backwards in the array.
You can lose...
1&4 - Like 0+1 you still have the original Disk 2 and the Mirror of Disk 1 just in a different order.
So it actually looks like the RAID 0+1 that the Modular Server is calling RAID10 may be more resilient than the common RAID10 as it may be able to absorb 4 combinations of two drives failing vs. 3.