[DRBD-user] io access very slow when in pause-sync

Denis Cardon denis.cardon at tranquil-it-systems.fr
Fri Dec 17 14:42:48 CET 2010

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


Hi everyone,

I had a strange behavior today on a drbd cluster. Yesterday I had to 
upgrade the slave peer, which went fine. I re-installed a debian lenny 
and the distribution drbd package (version drbd8-modules-2.6.26-2-amd64 
2.6.26+8.0.14-6+lenny3, same version on master peer). Once I started 
drbd service, the sync started fine.

Because the VMs when high on iowait, I switched to pause-sync. However 
the io throughput was still crawling at few MBps, with high iowait on VMs.

On the master drbd node, the load was high while the iowait was null. 
Idle was at 98%... Here is the top header :

=======
top - 10:49:23 up 1 day, 15:27,  2 users,  load average: 14.61, 14.39, 11.46
Tasks: 265 total,   1 running, 264 sleeping,   0 stopped,   0 zombie
Cpu(s):  0.0%us,  1.3%sy,  0.0%ni, 97.6%id,  0.0%wa,  0.3%hi,  0.8%si, 
0.0%st
Mem:  12337136k total,  1412376k used, 10924760k free,   357576k buffers
Swap:   524280k total,        0k used,   524280k free,   833320k cached
=======	

Finaly I shutdown drbd services on slave node, and everything went back 
to normal iowait/throughput wise. Now the master node is WFConnection.

So I wanted to now if I did something wrong to get such bad performance 
on pause-sync.

Here is my /etc/drbd.conf :
========
resource r0 {
	protocol C;
       on srvsan1.pnrvf.local {
       device /dev/drbd0;
       disk /dev/sda3;
       address 192.168.234.1:7788;
       meta-disk internal;
}
	on srvsan2.pnrvf.local {
	device /dev/drbd0;
	disk /dev/sda2;
	address 192.168.234.2:7788;
	meta-disk internal;
        }
	syncer {
        rate       700000;
     }
startup {
degr-wfc-timeout 15;
}
}
========

/dev/drbd0 is based on a primary partition. /dev/drbd0 is sliced up in 
lvm which are exported with iscsi-target to xen servers. It has been 
running for 3 years without problems (with upgrade in between).

So I was really puzzled with the behavior of the pause-sync. As I didn't 
found anything on the internet, I would be very glad if someone could 
point me to some hint.

Cheers,

Denis


-- 
Denis Cardon
Tranquil IT Systems
44 bvd des pas enchantés
44230 Saint Sébastien sur Loire
tel : +33 (0) 2.40.97.57.57
http://www.tranquil-it-systems.fr




More information about the drbd-user mailing list