View Full Version : Quantian clustering

03-02-2005, 01:47 PM
Not sure if this is the right forum for this but I'll give it a shot and prepare to be moved

So I am building a fairly large cluster (~60 node) for the quantum computing lab at my school... and I was hoping to get some help with some of the configurations that I am struggling with... The requirements for the cluster are: 1) the IPs for the nodes are statically assigned based on MAC... 2) the DVD be copied to the HD and used to boot the nodes (this should be easy but for some reason I am struggling with it)... 3) the hard drive should be shared over the network so that work done on the cluster can be stored there and accessed via SSH... any thoughts would be most appreciated and thanks in advance =)

PS... well, it looks like fixed mapping is achieved via the PXE config on the server... editting the /etc/dhcpd.conf file... but openmosix overwrites this file when it starts... intriguing problems...

03-04-2005, 09:09 AM
Do search for ClusterKnoppix, Quantain is based on it.
1. I have come to the same conclusion about the dhcp being overwritten. It seems like a good way to bind the MAC to a IP address for a diskless node. I am currently booting nodes from a short script. It uses etherwake+MAC and sleep command to boot the nodes in a timed orderly fashion. Along with the Terminal server allocating ip address in a decending manner. Default ip is, next, ect. It works by giving the same IP to the same MAC address. It's not great but it works. I have also have had NFS mount get "swamped" by to many nodes booting at once. Some of the nodes will wind up as zombies...
I also shut them down with a script. To boot again correctly, they need to be powered down to a "warm" state. The command that works for me is "ssh NODE halt -p -d -i -f"
2. Many have struggled with a hard drive install. There is an * missing from the Knoppix-terminalopenmosix script. Search this forum for a more detailed answer.
3. I was able to shorten the node boot times considerably :
create a file with this name and copy the pxelinux.default to it. Place it in the pxelinux directory. Now the node wont nibble slowly away at the ip address it skips straight to default. While this is good it dosent bind it to a MAC address as far as I can tell.

03-07-2005, 07:29 PM
thank you for the ideas =)
i switched the cluster over to clusterKnoppix since it really doesn't matter... I was able to get the HD to hold the boot image for the nodes... and I am intrigued by the idea of powering them on in order like you mentioned... what sort of script was it (if you don't mind sharing) that enabled you to do that?
I also am a bit confused about the procedure to shorten boot time... I understand what you are getting at with the creating of a file so the node doesn't try to trim away at the address... is the file name just those 8 characters? thank you again for your help

03-18-2005, 05:05 PM
I will post a pretty good how-to soon.
sorry my time is short at the moment