« How to use a watchdog timer | Main | The cost of un-availability - and the value of a bad example »

10 October 2007


Feed You can follow this conversation by subscribing to the comment feed for this post.


I don't totally agreed that SCSI-reserved disk is quorum disk. in redhat and m$ cluster suits quorum disk is shared partition (FC or SCSI only, I think) which writes by cluster members, each to it's own "sector". successful write is count as vote.

Alan Robertson

Not all SCSI-reserve disks are quorum disks, and not all uses of shared disks are quorum disks.

In the situation you're describing, I wouldn't call a disk used this way a quorum disk. It's being used as a communication method - a method of sending "I'm alive" or heartbeat messages. Used in this way, the disk is another communication method - a disk-based networking scheme - and when it comes to communication paths - the more the merrier.

Adding communication methods doesn't eliminate the need for quorum, nor the need for fencing, and it doesn't eliminate split brain. For example, there was a scheduler bug in Red Hat 2.4.18-2.4.20 which caused Heartbeat to stop running for nearly an hour. This would make the other node think that node was dead, but it wasn't. And, since it wasn't running, it wouldn't notice that it didn't have quorum.

Nils Goroll

You write: "As of this writing, I believe that the quorum daemon feature is unique to Linux-HA."
Sun Cluster supports Quorum Servers (scqsd) sice Version 3.2. Docs link: http://docs.sun.com/app/docs/doc/819-5360/gbdud?a=view

Alan R.

Thanks for catching this problem. I'll correct the text. After writing this, I also found out that HP had a similar mechanism. I'll update both this page and the one here: http://techthoughts.typepad.com/managing_computers/2007/10/more-about-quor.html with your links. Thanks!

The comments to this entry are closed.

Become a Fan

AddThis Social Bookmark Button
Blog powered by Typepad