[DRBD-user] Periodic (200 seconds) io peaks with DRBD 0.7.22

[Joschi Brauchle]

Hello everyone,

first of all, I know DRBD 0.7.22 is WAY outdated! Still, this is the 
only Novell-supported version for SLES10SP3 currently. Thus I would like 
to solve the problem with this version if possible somehow... :-)

So the problem we have is the following:
There are periodic IO load spikes that are causing the system to 
seriously hang for a couple of seconds.

We have been running 'sar' for a couple of days and just now discovered 
that the peaks disappear when we stop the DRBD service. There are 
exactly 18 peaks an hour, which make for a period of exactly 200 seconds 
between the peaks. The peaks happen on BOTH, primary and secondary node. 
So we stopped DRBD on the secondary node and the peaks went away there.

Unfortunately the 2.6.16 kernel of SLES10SP3 does not allow for process 
wise IO statistics, e.g. the program 'iotop' is not available. Hence we 
only have IO stats for the whole system.

Please find a graph from 'isag' attached which shows the periodic peaks.

This is our very basic drbd.conf:
--------
resource r0 {
   protocol C;
   on pollux {
     device     /dev/drbd0;
     disk       /dev/sda1;
     address    10.0.0.1:7788;
     meta-disk  internal;
     }
   on castor {
     device     /dev/drbd0;
     disk       /dev/sda1;
     address    10.0.0.2:7788;
     meta-disk  internal;
     }
   syncer {
     rate 24M;
   }
--------

I would be grateful for any help with 0.7.22 :-)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: isag-snapshot.png
Type: image/png
Size: 28340 bytes
Desc: not available
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20110627/ea238cfb/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5996 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20110627/ea238cfb/attachment.bin>