raid 0+1

[John Kufrovich]

Basil,
Your first sentence isn't correct.

RAID 01 can only handle two disk failures. One in each mirrored set.

RAID 10, can handle multiple disk failures as long as a failure isn't associated with an already failed mirrored member.

[Basil Vizgin]

John,
RAID0+1 of 10 disks (D):
---
DDDDD mirrored to DDDDD

STRIPE mirrored to STRIPE
---
Array lost one disk (X-failed drive), so secong set failed:
---
DDDDD mirrored to DDDXD
so
STRIPE mirrored to FAILED_STRIPE
---
But, as I understand, disks in failed stripe is still rotating and also can fail, for example:
---
DDDDD mirrored to DXDXD
so we have again
STRIPE mirrored to FAILED_STRIPE
---

Therefore in theory we can loose up to all disks of FAILED_SET.
Of course, workload in survived set is greater, so probability that next failed drive goes to working set is very hight.

[John Kufrovich]

Basil,
Your explanation is closer to a RAID 1+0 than 0+1.

Let's look at it in a different way.
With RAID 0+1, you have a
RAID 0 mirrored to a RAID 0.
What happens to a RAID 0 LUN when a drive is faulty? You loose that LUN. Same thing here, you loose that mirrored set.

With RAID 1+0, using 8 disk.
Drives 1 and 5 are mirrored (RAID1)
Drives 2 and 6 are mirrored (RAID1)
Drives 3 and 7 are mirrored (RAID1)
Drives 4 and 8 are mirrored (RAID1)
Then your data is stripped.
With this scenario, we could loose a single drive from each mirrored (RAID1)set and still function. In this case we can loose a max of 4.

[Basil Vizgin]

John,
In you example you assuming that next disk can fail only in active (survived) RAID0 set of RAID0+1, but by theory of probability it can fail in already failed RAID0.

[John Kufrovich]

You are correct, you could have a second failure in the failed RAID0. BUT, a failure is more likely to occur during I/O operations. The failed RAID0 will or should be marked BAD and no further I/O will occur.