Thread: fragmention issue with ext4: e4defrag?

fragmention issue with ext4: e4defrag?

From
Marc Mamin
Date:
Hello,

Has anyone some experience using defragmentation tools on Linux against tablespaces ?

we are facing fragmentation problems with postgres instances having a few TB of data.
( RAID 5 )

I/O througput decreased from 300MB/s to 160.


- We first moved some schemas to separate servers.
  After that we still have  150'000 tables in 1.5 TB

- Now we are in the process of vacuuming FULL historical tables which are not written anymore.
  This seems to improve the I/O considerably

Our remaining issue is that the free space fragmentíon is still suboptimal
so that fragmention will probably start again soon.

Would it make sense to use a tool like e4defrag
(http://www.linux.org/threads/online-defragmentation.4121/)
in order to defrag the free space ?
And how safe is it to use such a tool against a running postgres instance?



many thanks,

Marc Mamin


Re: fragmention issue with ext4: e4defrag?

From
Kevin Grittner
Date:
Marc Mamin <M.Mamin@intershop.de> wrote:

> I/O througput decreased from 300MB/s to 160.

I don't have any experience with ext4 defrag tools, but just wanted
to point out that the difference in performance you cite above is
about the same as the difference between accessing data on the
outer (and usually first-filled) tracks on a disk drive and the
inner tracks.  One of the reasons performance falls as a drive
fills is that the OS is compelled to use slower and slower portions
of the disk.  Part of the benefit you are seeing might be due to
freeing "fast" tracks and data being relocated there.

--
Kevin Grittner
EDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company