[DRBD-user] slow write performance

Arnold Krille arnold at arnoldarts.de
Wed Sep 12 21:53:10 CEST 2012

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On Wed, 12 Sep 2012 09:42:47 -0400 "Dan Barker"
<dbarker at visioncomm.net> wrote:
> a) use oflag=direct or dd will just test caching.

Especially if you only copy 11MB. Use oflag=direct and copy much more

And then a tip from some very practical experience: Test with a
filesystem-benchmark like dbench. dd only tests sequential read/write,
which isn't really of any practical use case unless you want to write
and read large movies. Everything else from recording DAWs to databases
to office usage is all random-access and fast reads are not helping
when you really need low seek-time-induced latency.

And that seek-time latency of drbd gets more important when you use
internal meta-data because then every write has to be marked in the
meta-data at then end of the partition. When placing the external
meta-data on a different disk (maybe even on an ssd next to the
super-fast-db-vm-image), we see an improvment of factor 2 in dbench's
results for IO-rate.

Have fun,

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20120912/6cd37b85/attachment.pgp>

More information about the drbd-user mailing list