My current preliminary conclusions on this box / workload:
- running psync is much better than sync
So you likely have a convincing case for Postgres guys to switch over to
pread/pwrite.
I will approach them, but I want to make sure I did all my homework first.
One question that bugs me:
the difference in performance between sync and psync engines only
surface with MD, _not_ when running over individual devices.
---
I ran Linux perf with these results:
https://github.com/oberstet/scratchbox/blob/master/cruncher/sync-engines-perf/individual-nvmes-sync.md
https://github.com/oberstet/scratchbox/blob/master/cruncher/sync-engines-perf/individual-nvmes-psync.md
https://github.com/oberstet/scratchbox/blob/master/cruncher/sync-engines-perf/md-nvmes-sync.md
https://github.com/oberstet/scratchbox/blob/master/cruncher/sync-engines-perf/md-nvmes-psync.md
---
md-nvmes-sync shows the "issue":
Overhead Command Shared Object Symbol
73.48% fio [kernel.kallsyms] [k] osq_lock
So while I think it would be good in general if PostgreSQL used
pread/pwrite instead of lseek/read/write when available, I am afraid
there might be a bottleneck in MD.
What do you think?
And if so, where should I raise this rgd MD? I have no clue where the
hackers of MD hang out ..
Cheers,
/Tobias
--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
