This post enticed me to do a little research on zfs compression, which
lead me to some interesting info an a question:
https://indico.fnal.gov/event/16264/contributions/36466/attachments/22610/28037/Zstd__LZ4.pdf
TLDR: A couple really interesting bits:
On their test data set, zstd-2 (4.8) beat out other compression ratios
soundly for performance with minimal difference in the compression ratio.
lz4-1 (145 MB/s) beat out lz4-9 (45 MB/s) on speed with no difference in
the compression ratio with their data set.
Question:
zfs was written to optimize usage largely for spinning rust.
What would be different in a version of zfs optimized for NVMe or SSDs?
How much can you tune zfs to optimize it for ssds?
Thanks
On 9/8/24 01:50, Jorge Schrauwen via illumos-discuss wrote:
Since I am on mobile I’ll top post…
Although some of the new features are indeed enticing, I value lack of dataloss
and stability more.
Once data has been lost, all trust is gone.
Work is a linux shop and most of my colleagues don’t care for ZFS because they
tried it and had all sorts of issues/loss of data. They rather use XFS or ext4.
(They don’t care much for btrfs either for the same reason, and prefer LVM to
fill the snapshot/zvol gap.)
There is probably a fair share of users that chose illumos for those values
that align closer with theirs.
I’m sure illumos loses some of the ‘new’ users that try out illumos because of
some of these choices of stability, don’t break master,… those users values may
closer align with they way openzfs does things, and that is fine. They have the
option to use it and seem to do so.
Ultimately picking an OS is always a mix of features and values the project has.
I think illumos is uniquely positioned here.
~ sjorge
On 7 Sep 2024, at 23:39, Gea <g...@napp-it.org> wrote:
Thank you for bringing the discussion forward..
Have your cake...
This is a term used by the Brits on Brexit in Europe.
While BS it basically says look for a way to get something new from
controversial demands.
As you say a project must have a focus on certain points be it stability or
compatibility and yes stability has priority but in the end stability is
nothing without compatibility when there are not enough users to justify the
whole effort beside a single company use case. This is not different to the end
of Sun that lost focus on users. Dying virtuously is not a serious option for
any OSS project.
From a user view, I do not see how the current Illumos follows Open-ZFS
development or see that important new Open-ZFS features find their way to
Illumos in time, so it cannot be more of the same.
As I am not a Illumos dev this is my personal outside view. There may be other
solutions from inside view.
Gea
g...@napp-it.org
-------------------------
------------------------------------------
illumos: illumos-discuss
Permalink:
https://illumos.topicbox.com/groups/discuss/T627f77e1b29a7b53-M69804f13feba105e940a7d98
Delivery options: https://illumos.topicbox.com/groups/discuss/subscription
------------------------------------------
illumos: illumos-discuss
Permalink:
https://illumos.topicbox.com/groups/discuss/T627f77e1b29a7b53-Mc907e2babca0f04733813e22
Delivery options: https://illumos.topicbox.com/groups/discuss/subscription
