Linux-HA Clustering ... Anyone use this before?

[DaveLembke]

Has anyone here used the Linux-HA clustering to see how well it works?

For my corporate application it probably makes more sense to replace one of our older dual xeon servers than place 2 of them into a Linux-HA Cluster for processing to load balance and extend their useful life.

The business has a tight budget, needs more processing power, and we have a spare same model 2U HP Dual Xeon server that is a switch over for when the primary goes down. My idea is to light up this other 2U server and configure them as a Linux-HA Cluster. The current OS is Fedora 17. So for the fact that everything is already running in Linux environment I am thinking this might be a direction to go, whereas if you were dealing with Microsoft/Linux-HA mix that would be a nightmare if its even possible.

The Linux-HA is also appealing because of redundancy if 1 of 2 servers die, the other keeps everything alive. The other alternative is to split up some of the processing between the 2 servers and have say accounting running on one and other processing on the other to get rid of the lag that has begun in the processing of the users needs. BUT this then means that if one of the 2 goes down hard, you dont end up back to slower pre Linux-HA performance at 1/2 the processing power, but instead you have a section of the business that wants to choke you if you dont fix it quickly so they can go home for the day. LOL

Thanks for all input on this one. If I do go with this Linux-HA project, I am going to set the original drives off to the side and install a new set of drives after copying over all the data that needs to migrate, so that if the Linux-HA project running into the late night early AM is a bust at 5am, I just need to install the drives back into the face of the server at the last minute and bring it back up live as if nothing ever happened..lol

[banjo67xxx]

Has anyone here used the Linux-HA clustering to see how well it works?

Yes, it works very well. I've done the class on it at work so I can tell you the easiest way to configure it is by installing ricci on the 2 F17 servers, then installing luci on another box where you want to administer it from. Luci just tells ricci what to install and its all setup.

The only thing that caught me out was the resource group dependancies for a service seem back to front to me.

Also, for a 2 node cluster you might want to define a quorum device and set up some carefully chosen heuristics to get the best node to win in a split-brain race. Which reminds me, never use manual fencing. Always use the ILO, DRAC or whatever to do proper power fencing.

[DaveLembke]

Thanks Banjo67xxx for your input on this. I have decided that I am going to take 2  spare workstations that I have and learn how to get this working before deciding to pull an all nighter if needed to cluster the 2 servers we have. If I run into any issues, I'll ask here.

Was nice to get a response from you on this as for I thought maybe no one had played with it here before and it would become one of those unanswered questions that fade into the past.. 

[banjo67xxx]

Was nice to get a response from you on this as for I thought maybe no one had played with it here before and it would become one of those unanswered questions that fade into the past.. 

I'm Red Hat Cluster certified and support it as my day job, so answering questions on Fedora cluster is no problem.