Conference turris::digital_unix

9327.0. "dataless high availabilty and bootp: help" by GRIFUN::BENVI () Fri Mar 28 1997 14:43

Hi all,
	I've been contacted  by a customer who need to setup a diskless
high available environment.
The configuration thet customer is trying to setup is roughly this:

	+--------+ scsi +------------+ scsi +------+
        | srv1   |------|shared disks|------| srv2 |
	+--+-----+      +------------+      +-+----+
           |                                  |
           |  ethernet                        |
          -+-------------+--------------------+---
                         |
                      +--+--+       
                      |     | diskless client
                      +-----+

Via DMS and diskless driver we can easily implement a single server
configuration. The problems arise when we try to add high availability features
replacing the single server with a couple of servers in a decsafe ASE
configuration.

We'd like that the dms server be the ASE alias of the NFS service conatining
the dms stuff, instead of the hostname (and address) of the server that
phisically boots the client.

We tried to setup a configuration installing the dms stuff in a shared disk of
decsafe configuration and setting up a NFS service pinting to that disk.

We hacked the fstab file in client configuration changing the mount point
of /usr file system specifying the service name  insetad of the server hostname.
We made simbolic links from 
/var/ase/<nfs-service-name> to the path containing the dms stuff

We hacked bootptab in both server specifying the rp field (that contains an
explicit reference to the server hostame) specifying 
<ase-service>:/clients/<client-name> instead of
<server-name>:/clients/<client-name>

We managed to boot the client from one of the two servers without problems.
Issuing "df" on the client we saw / and /usr filesystem nfs-mounted on
<ase-service>:/<something> as we expected.

We tried to relocate the service: the client hung at the fisrt command issued
(es: ls). 
We suspect that the client has somewhere harcoded the IP address of the server
that booted it instead of the ip address of the ase service.

Noe that root filesystem seems not to be mounted accessing fstab by the client,
but seems to be mounted during the boot process using informations
passed via network by the bootp/tftp server. Is it true?

I had a look to bootp sources and I found that the server IP address is written
in the packet sent to client. Is it used in any way by the client?

Can anybody explain how the remote boot process happens?
Any chances to make the above configuration to work?
If yes, how?

I read in a quite old note on ASE conference that what we'd like to do could be 
done with some patches to bootp and tftpd.
Can anybody explain how? (If customer would like to pay for this...)

HELP!
Thanks in advance
Angelo

P.S. crossposted on digital_unix and ase notesfiles.