whkr@ciba-geigy.ch (Hans Kraft) (03/22/91)
As we are setting up a small test network we need to know, how a HP9000/3xx (4xx) diskless workstation boots, especially how does it get its internet address. Is its first action to ask on the lan, if some server knows the workstations internet address? Or does it use ethernet addresses until....? Any help will be appreciated. Thank you. Hans Kraft --------------------------------------------------------- He is not too bad until you ask him to think. (S. Holmes)
milburn@me10.lbl.gov (John Milburn) (03/23/91)
whkr@ciba-geigy.ch (Hans Kraft) writes: >As we are setting up a small test network we need to >know, how a HP9000/3xx (4xx) diskless workstation >boots, especially how does it get its internet address. >Is its first action to ask on the lan, if some server >knows the workstations internet address? Or does it >use ethernet addresses until....? The discless machine broadcasts (ether address) a "boot me" message. The server has a daemon running which is configured to respond to such requests from certain ethernet addresses. The ether address is mapped to a hostname, and at boot time the IP number is assigned from the hostname mapping in /etc/hosts. Since the protocol is a simple ethernet broadcast, and therefore not routable, the server and the client must be on the same subnet. A sample entry from /etc/clusterconf: 08000903c6ab:2:me9:c:1:4 The fields are as follows (from `man 4 clusterconf`) machine ID The ETHERNET address of the attached LAN card. This is a 12 character hexadecimal number. cnode ID An integer between 1 and 255 inclusive. Used to identify cnodes within a cluster. Each entry in /etc/clusterconf must have a unique cnode ID. By convention, the cnode id of the root server is 1. cnode name The name associated with this cnode of the cluster. The cnode name may be up to 8 characters long. Each entry in /etc/clusterconf must have a unique cnode name. cnode type A single character. If this machine is the root server, the character will be 'r'; otherwise, it will be 'c'. swap location If this machine uses the root server's swap space, this will be the cnode ID of the root server. If swapping locally, it will be the cnode ID of itself. csp The default number of kernel level server processes to create when the csp(1M) command is executed. -jem -- John Milburn milburn@me10.lbl.gov (415) 486-6969 "I am successful because I am the only person in my city who is not heavily addicted to powerful narcotics." -Cerebus
darko@hpfcdc.HP.COM (David Arko) (03/27/91)
I probably don't know *everything* about the network boot process
but I will take a stab at your questions.
What happens during the boot process is:
1) When the diskless client tries to boot, it sends out a broadcast
boot request to see if any server will boot him.
2) If a server running rbootd(1M) has an entry in it's /etc/clusterconf
with that clients Ethernet address, (LLA), then it will respond to
the client saying that this server can boot him. Rbootd then looks
at the servers /etc/clusterconf file and translates the LLA into a
hostname, this hostname is now what that client will call itself
when it boots.
3) After rbootd has completed the boot request and the client is starting
up, it will run the /etc/netlinkrc script. This will run the
ifconfig command which will set up the internet address for the
client host by using its `hostname` value (remember that the
hostname was given by the clusterconf file). So it is at this point
that the internet address gets assigned to the client. Up till now
it was using the Ethernet address (LLA).
Hope this helps...
-- David Arko (darko@hpfcrn.hp.com)perry@hpfcdc.HP.COM (Perry Scott) (03/28/91)
More info is can be found with 'man 1m rbootd'. Perry