Printable View
Can anyone give some advice to a cluster nubie? i have looked for instructions on how to setup a simple cluster..... 2 pc's.
From CD not HD install.
1. I boot the first pc, run the Start KNOPPIX Openmosix Terminal Server.
2. Go through the setup, and start the server.
3. Boot the second pc via pxe, boots great, all works.
4. Open Openmosixview to only find one pc in the display window? It only shows the local machine not the other pc.
5. What am I missing?
Thanks in advance.
Ryan
The easiest way I found to do this is to burn several clusterKNOPPIX CDs and boot them on several PCs on a LAN. This creates several nodes that can interoperate and already see eachother. Just run openmosixviewer in a shell to see all your nodes (from any of the nodes). BTW, I can't run LTSP because of a Wimpdoze DHCP server on the same LAN....
To test the automagical load balancing of the openMOSIX kernel, run Kandel from one node and watch the processes auto-migrate to other nodes on the openmosixviewer. The first time I tried this with clusterKNOPPIX it actually did not work. I've been using Quantian that's based on clusterKNOPPIX lately and it works fine!
I've had up to 25 nodes each running a PIV 2.6 GHz with about 800MB RAM each. Thats like one PC running a 65GHz processor with 20GB RAM!
Good Luck,
AJG
I have two machines connected in a local LAN and the newer machine also has a ethernet card through a cable modem to the internet.
When I just booted them with the CD they didn't find each other.
Here's what I did - it might help you.
Boot both machines with the ClusterKnoppix CD.
On the new machine configure the local connection first (things go wrong if the cable modem is done first). I forget the precise IP number that is used by default but call it X.Y.Z.1
I accepted that default number and all the other defaults. Then I configured the cable modem connection (as DHCP).
Then as root execute:
omdiscd -i eth0
where eth0 (or eth1 etc.) is the name of the connection.
Now on the old machine I configured its ethernet card. The default IP number is again X.Y.Z.1 and I changed it to X.Y.Z.2, accepted all the other defaults and then executed:
omdiscd -i eth0
Now they could both see each other and processes would migrate and if you save the configuration on each machine and boot up with myconfig=scan they'll come up next time without needed to be configured.
It was really cool seeing the processes migrate. In 1982 I did my M.Sc thesis on process migration and it was nowhere as easy to do then.
Best Wishes
Pete
Oh, really? That sounds familiar. In 1984 I completed my Masters of Engineering at Cornell University where my final project had to do with process migration too! What a small world. All I remember about it was massive amounts of code in C....
Regards,
AJG
Thanks everyone, got it working. Is there a way to centralize storage space?
Thanks
Ryan
.... and a PDP-11/23 connected to a PDP-11/70 via a 9600 baud terminal line.Quote:
Originally Posted by A. Jorge Garcia
Those were the days :-)
I haven't played with it very much but the filesystem on each machine is visible to the others through a directory called (I think) /mfs under which is the processor ID (the last digit (N) of the IP number x.y.z.N). It contains directories /mfs/0, /mfs/1 etc.Quote:
Originally Posted by Ryan
I presume you could make node zero the one with all the disk space and then code your programs to access databases or whatever in /mfs/0/. When/if they are migrated they will all still be referencing the central disk space. There are probably better ways though.
Best Wishes
Pete
That would be PDP 11/780s....
Anyway, if you are using knoppix-installer to make a hdinstalled node to save your work, you need to set-up the mfs direectories by hand. See the clusterKNOPPIX site for more info as I forget the exact details. I have one such node based on QUANTIAN in my lab.
Enjoy,
AJG
Oh ! how can you revive "souvenirs" . I still have in my head theQuote:
Originally Posted by el_supremo
image of daniel Taupin (the guy who wrote 'almost alone' at the beginning the "algol68 compiler" )modifying the 1m2 switching
board of the huge printer-peripheral of the ibm650 filled with double triodes to be changed almost every 1O hours ...
10mhz fast transistors were Ge 1N1143,2N511...(mobility coefficient of course)and 3cm X band with Varian,Thomson klystrons !!!!That was in the sixties !!PDP did not exist yet !
No. It was a PDP-11/70. The 11/780 was a VAX - the first one I think.Quote:
Originally Posted by A. Jorge Garcia
I sorted punched cards for a LEO computer in England. MTBF was measured in hours - and not many of them either :-)Quote:
Originally Posted by wangji
Best Wishes
Pete
how do you boot a pc usinga network card that is pxe capable/
what exactly do you ahve to do in clusterknoppix?
I just recently tried to setup several nodes with clusterKNOPPIX 3.6 CDs and the cheatcode "knoppix toram." If I boot normally from CD (no cheats) or from hdd (after knoppix-installer + setting up mfs) openmosixviewer can see all the nodes and all the migration. If I use toram, I only see the local node, sometimes I see a remote node in red.... Is there a fix for this?
TIA,
AJG
can you be more specific about setting up hdd (after knoppix-installer + setting up mfs) openmosixviewer
what is "setting up mfs ? mean
thanks
peter
Take a look at:
http://www.knoppix.net/forum/viewtop...r+mfs&start=50
Regards,
AJG
for plener: first u make sure ur master node is up and can access the network, then u set up the knoppix openmosix terminal server as stated, then u make sure your client node's network card support pxe boot (u need to enable network boot in your BIOS), then just start ur client node.
and to: el_supremo, A. Jorge Garcia and wangji
I finally found you guys that used quantian successfully, do u mind give me a little hint about my problem?
I may start quantian in my vmware (boot:migshm), and set up the openmosix terminal server successfully, and then my node and PXE boot from it, my openmosixview also found 2 cpus now, however, while i just randomly open some applications to see if the load is being migrated and the other cpu is not showing any load at all.
and I found this error on the other node saying that:
openMosix: Auto-migration of local-processes from this node re-enabled
openMosix: guest-processes now accepted again
MFS already mounted!
omdiscd[2774]: bind failed on interface 0: Address already in use
omdiscd[2774]: Unable to initialize network. Exiting.
any comments are greatly appreciated
I've never gotten PXE to wprk right in my lab.
I recommend all nodes boot a CD or all nodes be hdinstalled. I've recently had more success with a CD only distro (no hdinstall option) that runs on a 200MB CD and, as a rexult, boots about as fast as hdinstall. This distro in BCCD (Bootable Cluster CD) which also uses openMOSIX. Maybe you sould try it!
Regards,
AJG
Thx Jorge:
The reason I used quantian PXE boot my slave nodes is that my slaves don't have DVD ROM, but anyways, I did progress a bit and see the load of the cpu shifting back and forth to different nodes using the IBM test script testapp.c here (thx to a friend who mentioned this in the early post)
http://www-128.ibm.com/developerwork...knop.html#main
I did try BCCD, and it appears not to support my netword card, so...what a pity.
It's real hard for the openmosix to utilize unless u use some big software such as blender or stuff...so openmosix is really got to develop much further.
Keep in touch, thx!