On volume creation the resiliency choices are:
- 2 or 3 way mirroring (= 2 or 3 replicas)
- single or dual parity (= erasure coding)
- Mirror-Accelerated Parity (= mirroring + erasure coding)
- Nested Resiliency (2-node only)
When choosing mirroring, each replica is placed on a separate physical node within the storage cluster. This means that 2-way mirroring requires a minimum of 2 nodes and 3-way mirroring requires a minimum of 3 nodes. 2-way mirroring most closely resembles RAID-1.
Mirroring provides the fastest possible reads and writes, with the least complexity, meaning the least latency and compute overhead.
Single parity keeps only one bitwise parity symbol, which provides protection against one failure at the same time. It most closely resembles RAID-5.
Dual parity implements Reed-Solomon error-correcting codes to keep two bitwise parity symbols, thereby providing protection against up to two failures at the same time. It most closely resembles RAID-6.
Parity encoding provides better storage efficiency than mirroring without compromising fault tolerance.
A volume can be part mirror and part parity. Based on the read/write activity, the new Resilient File System (ReFS) intelligently moves data between the two resiliency types in real-time to keep the most active data in the mirror part and the least active data in the parity part.
Mixed resiliency can be considered when most of the data on the volume is "cold" data, but some sustained write activity for some data is still expected.
Nested Resiliency (2-node only):
This resiliency enables to support two simultaneous failures. When using Nested two-way mirror, the data is copied 3 times across the cluster with 2 data instances per node as a result (equal to 4-copy mirror). Can be also used in Multi Tier with one tier using two-way mirroring and the other tier using RAID5 parity.
Microsoft, S2D [SDS]/2019, Datacenter
Disk Failure Protection, Reads/Writes, Data Availability