[DRBD-user] Slow disk performance during sync stage

[Tim Wood]

Hi,
I've noticed that the performance of my disk drops significantly
during the initial synchronization stage when I bring a second drbd
node online.  After the "Sync" period ends, performance is fine.  I am
running over a 100mbps link, and I have limited the sync rate to 4M.

Here is my config:

common {
  protocol A;
}
resource r0 {
  device        /dev/drbd1;
  disk          /dev/VolGroup00/drbd0;
  meta-disk internal;
  on rcs6 {
        address 192.168.246.6:7789;
  }
  on rcs7 {
        address 192.168.246.7:7789;
  }
  net {
  }
  syncer {
        csums-alg md5;
        rate 4M;
  }
}

I'm not sure why it would matter, but I am using the device as a disk
for a Xen virtual machine.  During the sync stage, small disk access
in the VM can take several seconds, and if I look at "top" it reports
most of the CPU time being spent as iowait.  Performance is erratic --
during part of the sync it will be fine, and at other times it will be
very slow.

Any idea if this is a problem with Xen interacting with DRBD in a
funny way, or if I just don't have my drbd setup properly?

thanks!
Tim