Searched the Site, but.....couldnt find a peep about clustering.
was wondering.....does clustering even EXIST for linux ???

I'm trying to build a HA email system, from 2 machines. (low-end)
I did hear bout Beowulf, but i couldnt find any references to service-specific clustering. (namely QMail,and its children)

Could any of the Gurus please shed some light here ??

Hope its not too hard a question I have asked (tis my first ).

Thanx,

nev_neo.
Linux Student.

Clustering is a VERY broad topic.

Beowulf clustering generally refers to a compuational cluster.

High Availability and load-balancing clusters are also quite common.

I use the "heartbeat" package to create a 2 node High Availability cluster of DHCP servers. Heartbeat can be used to make many services highly available though - web servers, DNS, e-mail, LDAP, etc.

This is basically how it works:

I have two nodes connected to the netwrok. Each network has a "real" ip address. There is also a "service" IP address that is associated with the active node. If that node goes down, or you force it to go to standby mode, it reliquinshes that "service" IP address and the second node becomes active. In my configuration, the two machines use a serial cable to send heartbeat infromation; this is how they know the status of the other node. More complex configurations are available, but you get the idea.

RedHat Advanced Server also has a cluster suite ( I believe that this provides load balancing and high availability.) If you need to do shared (or replicated) storage, things become considerabley more complicated.

If you don't want to pay for RedHat Advanced Server and want a "professional" cluster solution, try Scientific Linux - it's RedHat Enterprise Linux recompiled from source and it is free. Also, I believe that it comes with the Cluster Suite package.



see http://www.linux-ha.org

http://www.linuxquestions.org/questi...der=descending

Or try this for a live cd based linux cluster:
http://bofh.be/clusterknoppix/

One really good popular clustering system is openMosix.
It required knowledge of compiling a kernel though.

Here are some websites reguarding openMosix-

http://openmosix.sourceforge.net/
http://www.openmosixview.com/
http://www.ibiblio.org/pub/Linux/doc...penMosix-HOWTO
or if you want to download HTML version- http://www.ibiblio.org/pub/Linux/doc...TO-html.tar.gz

openMosix isn't really that hard to setup. I use it on my 2 node cluster, and when I used KDE I had openMosix view.

DOH !!!!
Was looking in the wrong section !!! Somehow thought it would be in Networking. Guess not.

Jimrt..... your setup is what i was looking for.
Thought i could have them on an Active-Active setup, but i've still got to work on that
Was looking to do some high level of load-sharing, but how would I be able to do that when I have only 2 systems---Total.
My understanding is...for this kind of setup, i would need to have a Master (head node), which distributes all incoming requests amoung the rest of the nodes.

Is there any other way ???

AM currently looking at Ultra-Monkey and LVS. Need to know if LVS would be possible in a 2 node Active -Active Setup ????

Thanx for the input.....looking forward for more.

There are software programs out there for doing load balcing (such as the appropriately named balance .) Also, there are allegedly network switches that can perform load balncing. I use round-robin DNS to balance the load between my OpenLDAP servers, but that probably is not a good option for a mail server.

The problem you are going to run into is data storage. If you use a shared storage array you need to be careful to avoid the "split brain" problem where two nodes actively try to use the storage device at the same time and end up screwing it up. Replicated storage - where each machine has its own seperate storage device ( ie.e a disk or even a parition) - and the updated data is "replicated" from one server to another is a lower cost option. This is not going to be a good option for an active active configuration though.

In my opinion, active-active should not be tried on any cluster where the data changes frequently; unless you are willing to invest a lot into hardware and software. A friend of mine is looking at doing an active-active load balacing e-mail system and they are purchasing a SAN as well as a hardware load balancer.

http://freshmeat.net/articles/view/458/ has some more information on high availability clusters and linug mag had a decent article last fall/winter.

Let me know if I can provide you with any more help. High availability clustering is very inetresting to me, but once you get into it you quickly realize that there are many tradeoffs to be made.