[DRBD-user] Poor fsync performance

Andy Dills andy at xecu.net
Thu Jan 10 04:40:25 CET 2013

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


On Tue, 8 Jan 2013, Andy Dills wrote:

> # lvcreate -n test -L 10G data
> # mkfs.ext3 /dev/data/test 
> # mount /dev/data/test /mnt
> # pveperf /mnt
> 
> CPU BOGOMIPS:      95995.32
> REGEX/SECOND:      838290
> HD SIZE:           9.84 GB (/dev/mapper/data-test)
> BUFFERED READS:    477.27 MB/sec
> AVERAGE SEEK TIME: 5.87 ms
> FSYNCS/SECOND:     631.64
> DNS EXT:           43.10 ms
> DNS INT:           1.18 ms 
> 
> The fsyncs/second should be tremendously higher.

For what it's worth, here's what pveperf looks like with the resource 
disconnected:

# pveperf /mnt
CPU BOGOMIPS:      95997.00
REGEX/SECOND:      877865
HD SIZE:           9.84 GB (/dev/mapper/data-test)
BUFFERED READS:    465.67 MB/sec
AVERAGE SEEK TIME: 5.83 ms
FSYNCS/SECOND:     1691.68
DNS EXT:           42.45 ms
DNS INT:           1.01 ms

And here are some simple DD tests:

Connected:

# dd if=/dev/zero of=/mnt/file.tmp bs=512M count=1 oflag=direct
1+0 records in
1+0 records out
536870912 bytes (537 MB) copied, 3.2879 s, 163 MB/s

Disconnected:

# dd if=/dev/zero of=/mnt/file.tmp bs=512M count=1 oflag=direct
1+0 records in
1+0 records out
536870912 bytes (537 MB) copied, 1.54669 s, 347 MB/s


Any ideas why I'm seeing roughly 40% performance when my nodes are 
connected?

Thanks,
Andy

---
Andy Dills
Xecunet, Inc.
www.xecu.net
301-682-9972
---



More information about the drbd-user mailing list