"dax" mount option for non-dax Db/App

19 views
Skip to first unread message

Anton Gavriliuk

unread,
Mar 26, 2019, 2:19:12 PM3/26/19
to pmem
If Db/App operates in non-dax mode, but uses fs direct I/O by setting O_DIRECT flag or in any other way, should I choose fsdax mode and mount fs with "dax" mount option ?, Or other words - "dax" mount option provides any advantages for non-dax Db/App which uses fs direct I/O ?

Anton

Andy Rudoff

unread,
Mar 26, 2019, 2:32:26 PM3/26/19
to pmem
Nothing beats gathering benchmarks results to decide on the right configuration for you.  But in general, you can think of fs DAX as equivalent to O_DIRECT on a non-DAX file system where sectoring is not used.  Of course, with O_DIRECT, the application is making specific choices on when the page cache should be used and when it shouldn't.  With fs DAX, the page is not used, regardless of what the application does.  Sometimes page caching is helpful, especially when the underlaying device is not as fast as DRAM.  And sometimes the page cache is unhelpful, because the application is managing the placement of data and wants full control.  That's why I say benchmarking is really the only sure way to tell which is better for a specific use case.

-andy

Anton Gavriliuk

unread,
Mar 26, 2019, 2:50:43 PM3/26/19
to Andy Rudoff, pmem
Ok, I just thought that someone already investigate that.  Now I checked it and got +20% IOPS at least for reads.  So fsdax mode and "dax" mount option should be used even not for non-dax Db/App which performs direct I/O.

Anton

вт, 26 мар. 2019 г. в 20:32, Andy Rudoff <an...@rudoff.com>:
--
You received this message because you are subscribed to the Google Groups "pmem" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pmem+uns...@googlegroups.com.
To post to this group, send email to pm...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pmem/5a919eb8-6769-4763-a606-5c1033a47376%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Anton Gavriliuk

unread,
Mar 26, 2019, 2:52:49 PM3/26/19
to Andy Rudoff, pmem
mistake....

even for non-dax Db/App which performs direct I/O.  

вт, 26 мар. 2019 г. в 20:50, Anton Gavriliuk <antos...@gmail.com>:
Reply all
Reply to author
Forward
0 new messages