Slashdot Mirror
ClusterKnoppix
chronicon writes "Knoppix is the ultimate live CD. No geek-kit should be without it. Now Wim Vandersmissen has taken it a step futher by adding openMosix functionality. Drop the clusterKnoppix CD in your "server", boot up... boot up some networked clients... Knoppix built in LTSP magic kicks in and ta-da--instant cluster!"
clusterKnoppix is in desperate need of mirrors. here's one (but i urge you all to make a .torrent or something):k noppix/
http://www.openmosixview.com/cluster
for a crappy yet less bloaty altenative, check out PlumpOS: http://plumpos.sourceforge.net/
clusterKNOPPIX_V3.2-2003-05-20-EN-cl1.iso.torrent
(also added to the main clusterknoppix website)
PXE describes a method where the NIC in the computer bootstraps the information it needs to boot off of the network. Many modern computers have NICs that support this. Newworld Macintoshes can netboot, as can most recent 3com cards (even my 3 year old Dell supports it).
Basically, the NIC makes a DHCP (or BOOTP) request for an IP address. The DHCP protocol allows the server to return the address of a TFTP (Trivial FTP) server along with the IP address for the client. The client contacts the TFTP server to get a kernel (vmlinuz), and then boots directly into that. From there, the kernel should be configured to mount its filesystems over NFS, and finish the boot process. I'm sure Google can point you to a more complete explanation.
What makes ClusterKnoppix so cool is that it's usually a huge pain to set up a TFTP/DHCP/NFS server correctly for multiple clients. ClusterKnoppix does it all for you, so all you need are some (really) "dumb" clients and all the heavy lifting is done for you.
> If you set this up correctly all the computers that you boot
> up with this become a mosix cluster?
an openMosix cluster, not a mosix cluster.
>Then all the users are terminals off of this cluster?
if you want, yes.
> So all of the users have some of all of the power of the
> Mosix cluster?
yes
> I just wonder how well mosix handles nodes dropping
> off and back on again.
if a node goes down for a small time, and then comes back, no problem. if a node goes down for a time long enough to finish his work, processes won't come back where they came from, so you (or your apps or scripts) have to take care of this situation. tipically in a cluster you don't want nodes to go down, never. this can be a situation tipical in a pc laboratory or the like, for an entire campus this probably is not adequate, you need something more "grid computing aware"
>Plus how well will can is scale?
it depends a lot on the speed of the connection between nodes, on the type and amount of traffic generated and so on the type of computation being made, on the number of nodes, on the speed of the clients, etc...
>Could you have five hundred or a thousand systems off in the cluster.
tecnically up to 65535 nodes (last 2 bytes of ipv4 address) if i'm not wrong. i was told biggest cluster of this types count 1-2k nodes, but i'm not sure.