There is a difference
Arc/L2Arc only caches datablocks on a read last/ read most optimazation,
not whole files and helps only on reads.
On a hybrid tiering pool with a fast special vdev you gain on read and
writes and with whole user selected files or large VMs
I agree that rule or filebased data tiering with a data move between
tier1 and tier2 storage is more complex than simple caching or the
default ZFS tiering based on data structures but let's look at worst case
A data move between tier1 and tier2 is an action like
increase recsize && rename file ex to file.tier && copy file.tier to
file && delete file.tier
This means every action must succeed for the next in the chain to happen.
If something happens during copy the final delete of the file.tier
should not happen so you have a renamed file.tier instead file. In worst
case use snaps. If the special vdev is full, new files land on regular
hd vdev/tier2 instead fast ssd/tier1. Its not nice that I see no method
to detect such files. Only solution is to force a manual move of such
files to tier1 when there is free space again on tier1 storage. A
problem could be open files without proper locking. If this is extremely
critical you may need to do tiering in a low busy time with shares
disabled.
This is not a use case for every user but for special use cased where
you want to move files transparently between fast/expensive tier1 and
slow/cheaper tier2 storage with paths remain intact. During the move
files are not accessable/locked.
Gea
On Sun, 12 Nov 2023 at 03:33, Gea <[email protected]> wrote:
Have i missed something?
This seems complex and potentially error prone. How is this superior
to using an L2ARC cache device to increase access speed to frequently
accessed data?
Cheers.
------------------------------------------
illumos: illumos-discuss
Permalink:
https://illumos.topicbox.com/groups/discuss/Ta9815f4d6c901308-M095380cc095dcb5a1446cd19
Delivery options: https://illumos.topicbox.com/groups/discuss/subscription
