Conference 7.286::fddi

1861.0. "Ethernet as backup (UNIX) - options?" by USCTR1::REICH () Wed Nov 08 1995 07:30


Our UNIX customer wants FDDI with the ability to quickly switch over to
Ethernet (TCP/IP). It does not have to be a true failover or even completely
automatic - just quick and easy. By way of example ...

  In the configuration drawn below, all of the traffic between the 4710 the
  two 2100 servers is going over ethernet. The customer wants it to go
  over FDDI and have the Ethernet service only the traffic from the VAX (GIMS).
  They also desire to keep the Ethernet connection on the 4710 to serve as a 
  backup to FDDI.

  What should the customer do in order to configure this so that 
  the bulk of the traffic between the 4720 and the two 2100 servers is
  going over FDDI?
 			+---------------+         +---------------+
			| Alpha 2100    |         | Alpha 2100    |
			| UNIX 2.0b     |         | UNIX 2.0b     |
			| node=sapserv1 |         | node=sapserv2 |
			+------+--------+         +------+--------+ 
                               |                         |
			+------+-------------------------+------+
			|              FDDI ring                |
                        |                                       |
			+---------+-------------------+---------+
                                  |                   |
                          +-------+------+     +------+-------+	  +---------+
                          |     Zylan    |     | Alpha 4710   |   |         |
                          |              |     | UNIX 2.0b    |   | VAX/VMS |
                          +-------+------+     | node=landry  |   |  GIMS   |
                                  |            | node=sapdata1|   |         |
                                  |            +------+-------+   +----+----+
                 Ethernet         |                   |                |
          +-----------------------+-------------------+----------------+-------+

For the Alpha 4710 ...
 * /etc/rc.config
	HOSTNAME="landry"
	NUM_NETCONFIG="2"
	NETDEV_0="te0"	 # ethernet
	NETDEV_1="fta0"  # FDDI
	IF_CONFIG_0="156.24.80.10" netmask 255.255.255.0
	IF_CONFIG_1="156.24.80.72" netmask 255.255.255.0
 * /etc/hosts
	156.80.24.10	landry   # ethernet
	156.80.24.72    sapdata1 # FDDI
For the two Alpha 2100 servers ...
 * /etc/rc.config
        NUM_NETCONFIG="1"
	NETDEV_0="fta0"
	IF_CONFIG_0="156.24.80.xx" netmask 255.255.255.0
 * /etc/hosts
        # setup the same as the Alpha 4710

In order to get most of the traffic on FDDI, we are considering the
following:

 Change the entries in /etc/hosts from ...
	156.80.24.10	landry   # ethernet
	156.80.24.72    sapdata1 # FDDI
 to ..
	156.80.24.10	sapdata1 # ethernet
	156.80.24.72    landry   # FDDI


This may help to get the traffic between the 2100s and 4710 onto the
FDDI ring, but it will make the customer's FDDI->Ethernet switchover
perhaps not so easy. We would have to change all of the "/etc/hosts"
files, but more troublesome would be the changes that we would have to
make in the setup of the primary applications; especially SAP R/3.

    This has nothing to do with FDDI.
    
    The Internet protocol suite doesn't address complex issues such as
    balencing between performance, availability, simplicity, (pick two),
    so it doesn't do anything.  Except give every interface an address.
    So, its simple, assign the addresses to each interface, and then
    design your applications to use the interfaces and paths that do
    what you want.  If you want failover or whatever, design your
    application to provide failover using the explicit interface addresses
    and your explicit knowledge of the network topology.