Browse Prior Art Database

Transparent addressing mechanism in a redundant network

IP.com Disclosure Number: IPCOM000010657D
Original Publication Date: 2003-Jan-07
Included in the Prior Art Database: 2003-Jan-07
Document File: 3 page(s) / 16K

Publishing Venue

IBM

Abstract

A method for transparently addressing multi-homed controller host in redundant networks is described. The transparency is considered of prime concern for applications that communicate among these controllers. The method defines a new form of virtual addresses that abstract from IP addresses of subject controllers. The method also applies to redundancy concepts for these controllers.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 48% of the total text.

Page 1 of 3

Transparent addressing mechanism in a redundant network

Introduction

A method for transparently addressing multi-homed controller host in redundant networks is described. The transparency is considered of prime concern for applications that communicate from controller to controller.

Definition 1: Cluster node.

A cluster node is a computer system, be it an stand-alone computer or even an embedded compute element, that feature more than one physical LAN interface. By implementation the physical packaging features more than one NIC (Network Interface Chip). A node can be a SC (System Controller), a BPI (Bulk Power Interface), or a logic cage (CEC or I/O).

Definition 2: Cages or groups of cluster nodes.

Cluster nodes can be grouped together into so-called cluster node groups or cages. This grouping may be dictated by common access to shared path's to special hardware, for instance.

Definition 3: Redundant network.

Cluster nodes as defined above can be used to implement redundant network paths by physically connecting the LAN interfaces of each cluster node to separate Ethernet hubs or switches.

Problem statement

    How can an addressing schema be devised for redundant networks composed of cluster nodes that hides he variance of the underlying network to application that communicate by means of the TCP/IP protocol suite? Or in other words, how can those applications address their counterparts without having to know details of the redundant path?

    We assume that application programs run on distinct cluster nodes. This means that the addressing mode must not preclude:

  The knowledge of the current state of the network paths. The selection of an individual path to address a peer node. Instead the application programs need only to provide:

The identity of the peer cluster node.

    We claim to have found a solution to the network addressing problem as stated above. The solution avoids the use of network addresses at all and introduces a new concept of a cluster node group or cage handle to achieve transparency.

Design for network interface redundancy

    In this chapter the concepts for addressing redundant network interfaces are carried forward towards a specification of addresses and their interpretation at runtime.

The IP-network configuration

    The redundant network of cluster nodes will be using the internet protocol (IP). The redundant network structure on the physical layer will be reflected as two separate IP-networks. Both physical networks will receive IP-addresses out off a reserved class A address space 9.x.x.x in order to prevent accidental

Page 2 of 3

misconfiguration in the customer network. The number of addresses needed is equivalent to two entire class C network addresses. This would be sufficient for 254 hosts per network.

    Note: In this example (and throughout the document) the following two sets of pseudo-class-C addresses 9.6.24 and 9.6.25 will be used.

The cage addressing interface (CAIF)

    In the chapter above the principle design of...