Hi Marcel, thanks for chiming in. > No. SMR won't be okay. There is no way to resilver a SMR disk in > reasonable time.
I understand that to be true if the SMR disk isn’t given “sufficient” idle time to defragment itself. What is “sufficient” has never been quantified, but since the drives are oriented to the notebook/desktop market, it seems that some read/write-profile that mimics that use-case would be okay - no? Assuming all logs are NOT written to an SMR-backed “cold” zpool (and the remaining writes are far and few between) and the reads are mostly covered by the ARC/L2ARC would that give the SMR-drives “sufficient” idle time? Regarding persistent L2ARC, I found it looks like it is in Illumos, per https://www.illumos.org/issues/3525 <https://www.illumos.org/issues/3525>, and thus presumably in r151036? That said, it seems that there are a number of improvements pending (per https://www.illumos.org/issues/13374 <https://www.illumos.org/issues/13374>), so maybe L2ARC isn't ready yet? Kent ------------------------------------------ illumos: illumos-discuss Permalink: https://illumos.topicbox.com/groups/discuss/T1d25146fb4ce0f44-Ma2293e07bc4a211fcaebe943 Delivery options: https://illumos.topicbox.com/groups/discuss/subscription
