Conference smurf::ase

    re: .1 (and 1871):  *ANY* address?  Class A or B or C?  Without regard
    to what else is on the network (whether intra- or internet)?  
    
    If I understand this correctly then I can, for example... 
    
        - have 2 cluster members with "real" IP addresses of 16.60.48.5 & .6
    		(we'll call them node1.wro.dec.com, node2.wro.dec.com)
    	- assign 1.2.3 as my subnet for MC interfaces
    	- assign 1.2.3.5 & .6 as my MC interface addresses 
    		(mc_node1, mc_node2)
    	- want to communicate via the Internet with the "real" 
    		nodes 1.2.3.5 or 1.2.3.6 (as registered by the NIC)
    
    The systems will not confuse/conflict MEMORY CHANNEL traffic involving
    mc_node1, mc_node2 & cluster_cnx (1.2.3.42) with traffic between node1
    or node2 and the "real" nodes 1.2.3.5, 1.2.3.6 and 1.2.3.42 (via FDDI,
    Ethernet, etc.)  
    
    Is this true?  Could I have used 16.60.48 as my MC subnet (same as my
    "real" subnet) as well?
    
    Ron 

    
    I asked about this back in October and never got a real good answer
    from any network guru's.  I'd really love to see an answer to Ron's
    question.
    
    .1 is not clear, to me, from any TCR documentation.  It really sounds
    like you need a whole subnet for yourself, and I'm constantly telling
    customers to use 10.0.0.n and I don't have a reason why that's okay
    (except that everybody does it).
    
    Anybody have an answer for Ron?
    
    - Sean
    

Hi,

My understanding of the addressing of the Memory channel network is the
following:

     1) This network must be PRIVATE to the cluster.

	This means the CLUSTER members must know each other via this network
	for their own (private) usage and that the OTHER systems on the 
	Ethernet or FDDI network should not access the CLUSTER members via
	this network. Thus, the SAME SUBNET can ALSO exist on the EThernet
	or FDDI network provided that NO CLUSTER member has to access that
	SUBNET. Thus, a big organisation which is in short supply of SUBNET
	networks can use that same SUBNET outside of the CLUSTER.

     2) This network can NOT have the same address as the Ethernet or FDDI
	network.

	This seams obvious to me, but I already have seen a lot of examples
	in this notefile with that mistake. As the Memory Channel is a 
	different interface than the Ethernet or FDDI one, it MUST have a 
	different SUBNET address. Otherwyse, the internal routing algorythm
	can not differentiate between the Ethernet and the Memory Channel 
	networks.

Can this be confirmed by Engineering ?

Regards, Manu.

    Thanks for the info, Manu.  Can anybody in engineering confirm this --
    particularly with regard to having an MC network number that is the
    same as one on the real net that you would want to talk to?  Can more
    than one cluster use this same network number?  
    
    I have a customer who is waiting for this answer before he installs
    TruCluster  (and has been waiting since Feb 07).
    
    Thanks and regards,
    
    Ron       

                     <<< SMURF::USERA:[NOTES]ASE.NOTE;1 >>>
                                    -< ase >-
================================================================================
Note 1871.2     *Unique* subnet required for MC interface in TCR?         2 of 3
NETRIX::"[email protected]" "williams@wast" 67 lines  28-FEB-1997 10:31
              -< Selecting a subnet for the MC physical network >-
--------------------------------------------------------------------------------
Why does the TruCluster product require a separate
subnet for the MC, memory channel, network?

        There are several answers to this question
        The cluster memory channel is a separate and
        distinct physical network on which only cluster
        members reside. If you attempt to use/share another
        network subnet address with the TruCLuster MC subnet
        you will run into the following problems:

        1. The current RIP protocol does not support routing of
        broadcasts across a subnet which spans multiple physical
        networks. This will break any networked application which
        utilizes broadcasts.

        2. Since at least two cluster members must be acting as
        gateways to provide application availability there would be
        multiple gateways interconnecting two physical networks
        within the same subnet. This condition is not supported
        by the current proxy ARP protocol.

        3. The traffic on the MC physical network needs to
        be tightly controlled to prevent saturation
        from outside sources.

        4. The traffic on the MC physical network needs to
        be contained such that the cluster does not
        saturate an outside subnet with its internal
        subnet traffic.

        5. Subnet based security with ifaccess.conf is used to
        prevent an outside node from spoofing or impersonating
        a cluster member.
What subnet address can I use for my cluster?

        The best answer to this question is to use your local
        network configuration and administrative methods and
        allocate a subnet according to your current model of
        subnet allocation. If you have no such method or if
        subnet addresses are a scarce commodity you must be a
        bit more creative.

        NOTE: IP addresses must be managed carefully to prevent
                colisions of the address space.

        It is important that you choose a subnet address which
        is not already in use. The Internet Assigned Numbers Authority
        (IANA) has reserved, in RFC 1918, the following three blocks
        of the IP address space for private internets:

                10.0.0.0        -   10.255.255.255  (10/8 prefix)
                172.16.0.0      -   172.31.255.255  (172.16/12 prefix)
                192.168.0.0     -   192.168.255.255 (192.168/16 prefix)

        It is also important to choose addresses and netmasks that are
        suitably subnetted to match any already existing subnets or
        networks in your distributed environment. In general this means
        to be sure to use a compatible broadcast mask with respect
        to the other subnets in your network.

        Use commands like "nslookup" to check if an IP address is 
	already in use.

        NOTE: RFC 1918:  http://andrew2.andrew.cmu.edu/rfc/rfc1918.html


[Posted by WWW Notes gateway]