2 ms·

I had a question about the mention of this under the RAID section: >"ZFS is not affected by the RAID-5 write hole." The "write hole" can exists in other RAID

• secabeen 9y ago
  All supported ZFS configurations. Copy-on-write means that the metadata for the new blocks is not written until all of the new blocks are committed.
• Dylan16807 9y ago
  ZFS doesn't use actual RAID. It uses independent writes to each disk, so there are no write holes possible.
  • bogomipz 9y ago
    But that's exactly what RAID is - independent writes to each disk. ZFS's stripe-mirror vdev is mirrored pair with a stripe written across those mirrors. How is that not actual RAID? Is that not RAID 10 with the ZFS addition of checksumming?
    • Dylan16807 9y ago
      True RAID depends on not-at-all-independent writes to each disk. The whole thing needs to be atomic to be safe, which is impossible without a battery backup. With ZFS each write to each disk is independently atomic, which is what avoids the write hole. You always know if the write succeeded on each disk. But I was to some extent conflating the different redundancy methods ZFS has. Some of them use the same sectors on each disk (I think) but some definitely don't. The writes are completely independent of each other under some settings.