Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Lars Ellenberg wrote:
>
> / 2004-07-06 10:21:44 -0700
> \ Sean Dwyer:
> > I have a seven disk raid array that is split into six partitions, and
> > I want to mirror each partition with drbd. I see that drbd.conf
> > discourages multiple drbd devices on one spindle, but how about on one
> > raid array?
> > Is performance likely to be an issue?
>
> test it :)
>
> > I am using drbd 0.6.12.
>
> there is the "sync-group" parameter, too,
> to serialize the synchronization of devices.
Sean,
I have a Raid 5 array with 5 disks and several partitions, and Lars answered
my query the same.
When I followed Lars advice I found that using the sync-group mitigated the
problems.
I have tried setting 2 of my logical devices in the same sync group, and found
this is a bad idea, it looked like the net through put was overall a bit
higher and more constant, but it took longer for the sync to occur.
My experience is put each partition of the a single physical raid array in a
separate sync-group and things will go ok.
It would be nice if the comment in the example 0.6.12 drbd.conf example
indicated something better like:
#echo "*****************" ; diff -u drbd.conf.orig drbd.conf ; echo
"***********"
*****************
--- drbd.conf.orig 2004-07-06 13:33:13.000000000 -0500
+++ drbd.conf 2004-07-06 13:40:32.000000000 -0500
@@ -156,6 +156,7 @@
ko-count = 4 # if some block send times out this many times,
# the peer is considered dead, even if it still
# answeres ping requests
+ sync-group = 0 #sync this disk first to reduce thrashing while
syncing.
}
on tcube1 {
@@ -179,6 +180,10 @@
protocol=C
fsckcmd=fsck -p -y
+ net {
+ sync-group = 1
+ }
+
on tcube1 {
device=/dev/nb1
@@ -187,6 +192,9 @@
# on the same spindle (hda)? performance will be bad. if you
# use several drbd devices, put them on different spindles;
# different channels/controllers won't be a bad idea for IDE.
+ # putting more than one drbd device on the same spindle kills
+ # resyncronization performance. this can be mitigated somewhat by
+ # putting each device in a different sync-group.
disk=/dev/hda7
address=10.1.1.31
***********
--
Todd Denniston
Crane Division, Naval Surface Warfare Center (NSWC Crane)
Harnessing the Power of Technology for the Warfighter