TUTORIAL Finally: The PIAF Cluster

[blanchae]

I came across the Kerrighed Linux cluster OS patch while looking for a simple solution for load-balancing. Interestingly on their Kerrighed cluster "hall of fame" page, UniMatrix has 60 nodes (Quad core CPUs) running Elastrix Asterisk PBX in a cluster called "The John Galt Line". That's 240 CPUs in total!

My understanding is that you would apply the kernel patch on the master and then install PiaF. All the slave nodes would use PXE to t*f*t*p to join the master and the slaves nodes use the master's kernel and file structure. The slaves are diskless. The Kerrighed cluster automatically load-balances among the nodes.

The current version 3.0 requires Linux kernel 2.6.30 - it is very kernel specific. There is a LVM version for evaluation. Now just to find the time to play with this...

 

[alesko]

Anybody tested this?

 

[blanchae]

Not so far. Tried on an atom based PC that I had but it was a waste of time. Once I return to work (on vacation) then I'll spend some time on it.

 

[blanchae]

Just did some more reading on the Kerrighed cluster, it doesn't look promising after all. Too experimental:

Kerrighed does not tolerate the sudden loss of a node. The system is not able to run correctly on other nodes afterwards.

So if you lose a node in the cluster - the cluster may crash. Not what I would look for in stability....

 

[tm1000]

http://hipbx.org/

 

[blanchae]

http://hipbx.org/

That is high availablity. A very neat concept using DRBD which is basically a virtual RAID-1 mirror using separate machines. It provides a neat solution for fail-over but only the active machine is participating. The rest are acting as "backups". It doesn't provide the distributed computing power that I'm considering.

In my simplestic world, I would like to see a PiaF distribution that if you needed to grow, you would turn on another server and it would PXIE boot to the original server, download an image and join/create the cluster. If one failed the other should take over like the high availability using DRBD and that would be sweet.

Something like Kerrighed meets DRBD.

 

[zane93]

http://linux4enterprises.blogspot.com/2014/02/two-node-pbx-in-flash-piaf-cluster.html

 

[wardmundy]

Great find, zane93. Thanks for sharing it.

 

[jeff.h]

Who is John Galt?