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



I've been reading your blog lately and I find it very helpful and informative. I'd be glad to see some article concerning project and load-balancing in general and how to achieve it using FOSS.

Keep up the good work!

Alan R.


Sounds like a reasonable idea - I can give an brief overview of the topics, but I'm not an LVS expert. Maybe I can get one of my LVS friends to write a guest blog article here for details. I'm also kind of jazzed about Cluster IPs at the moment. They're not as capable, but are much easier to set up.

In the meantime, you might check out Karl Kopper's "The Linux Enterprise Cluster" - which concentrates on LVS (and LVS with Heartbeat).

Reece Dike

You may not know the answer to this since you mainly deal with linux, but, I bet you know somebody that does.

How does Sun Cluster accomplish fencing? I am getting ready to implement a RHEL 5 cluster and it talks about using a ‘power controller’ and/or SAN switch fencing. When we were using Sun Cluster 3.1 we did not have either of those. I know we did have a quorum device and that sun wanted the quorum device to be able to do true SCSI-3 reservations. Why can’t Linux do the same thing? Or, Is sun ignoring a failure scenario that has a very low probability of occurring?

I have read your article entitled ‘Split-brain, Quorum, and Fencing – updated’, but, I don’t understand how sun is able to STONITH and/or fence a node without a ‘power controller’ and/or SAN switch fencing.

Alan R.

SCSI-3 reservations are a reasonable way to fence things. It does allow for multiple machines to be able to be fenced in at a time, which could be used to make sure all the interfaces were all able to get in.

With SCSI-II reservations exactly one server interface at a time can get in. But SCSI-III reservations are much more flexible (but I'm not an expert on them). However, SCSI-III reservations are not yet implemented correctly by every disk device. But, the key is this: It is third party (the disk device/controller), and it can be triggered without the cooperation of the node being fenced out. It _does_ only stop access to disks, so if you have other devices (or networks) that you need the hosts to be excluded from, you'll have to take care of them separately. So, for most purposes it will work well, because it's usually disks that you're most concerned about.

It's kind of a pain to set up correctly, but it should work fine for those systems that you've certified are compliant with your particular application. That's an additional kind of pain, but again, if you're Sun, and you designed the disks, and you designed the HA system, this isn't too hard to take care of.

The comments to this entry are closed.

Become a Fan

AddThis Social Bookmark Button
Blog powered by Typepad