bug? dsync extremely much slower on forward than reverse direction
I have been experimenting with dsync lately, and to my surprise I found that dsync in forward direction is extremely slow compared for reverse (-R) direction for the exact same data, disk caches flushed both times. I have created a docker-compose setup to prove my findings. Am I missing something here, or is this a bug?
link to tests: https://github.com/bjne/dovecot-dsync-benchmark
-- Bj(/)rnar
Here are some numbers from a test run:
dst: full pull sync from src 6m19s dst: delta pull sync from src 2s
src: full push dync to destination 4h14m34s src: delta push sync to destination 6s
So the initial sync is 4h14m (for push) vs 6m20s (for pull)
Can someone answer what is causing this very odd behavior?
man. 18. nov. 2024 kl. 13:18 skrev Bjørnar Ness <bjornar.ness@gmail.com>:
I have been experimenting with dsync lately, and to my surprise I found that dsync in forward direction is extremely slow compared for reverse (-R) direction for the exact same data, disk caches flushed both times. I have created a docker-compose setup to prove my findings. Am I missing something here, or is this a bug?
link to tests: https://github.com/bjne/dovecot-dsync-benchmark
-- Bj(/)rnar
-- Bj(/)rnar
It's probably fixed by https://github.com/dovecot/core/commit/3001c1410e72440e48c285890dd31625b7e12...
On 25. Nov 2024, at 14.51, Bjørnar Ness via dovecot <dovecot@dovecot.org> wrote:
Here are some numbers from a test run:
dst: full pull sync from src 6m19s dst: delta pull sync from src 2s
src: full push dync to destination 4h14m34s src: delta push sync to destination 6s
So the initial sync is 4h14m (for push) vs 6m20s (for pull)
Can someone answer what is causing this very odd behavior?
man. 18. nov. 2024 kl. 13:18 skrev Bjørnar Ness <bjornar.ness@gmail.com>:
I have been experimenting with dsync lately, and to my surprise I found that dsync in forward direction is extremely slow compared for reverse (-R) direction for the exact same data, disk caches flushed both times. I have created a docker-compose setup to prove my findings. Am I missing something here, or is this a bug?
link to tests: https://github.com/bjne/dovecot-dsync-benchmark
-- Bj(/)rnar
-- Bj(/)rnar
dovecot mailing list -- dovecot@dovecot.org To unsubscribe send an email to dovecot-leave@dovecot.org
participants (2)
-
Bjørnar Ness
-
Timo Sirainen