Cluster design documents
------------------------
$Id: 00README.txt,v 1.2 1999/12/16 21:15:22 sct Exp $

OVERVIEW OF THESE DOCUMENTS

This set of documents is not a comprehensive design spec for clustering.
Rather, it is a set of miscellaneous documents including both discussion
documents and work-in-progress design drafts, not for a whole clustering
system, but for a core set of APIs intended to provide a comprehensive
and robust infrastructure on top of which true clustering services can
be layered.

So, you won't find any proposals for IP takeover or for clustered
filesystems here.  You _will_ find proposals for APIs which will let the
IP failover manager communicate the state of the running IP interfaces
to the rest of the cluster, or to allow other services to be started and
stopped as appropriate if an IP address is migrated from one node to
another.  

The motivation for this work is primarily that although there are many
distinct clustering projects under way for Linux, there is no
general-purpose framework to provide solutions for some of the hard
problems such as quorum management, massive scalability and management
frameworks.  A successful outcome would be a set of core cluster
APIs which are both simple enough that arbitrary other (existing or
future) cluster services can take advantage of them easily; and powerful
enough that there is real benefit to be had from using them.


In this directory you will find the following documents:

  goals.txt:	Outlines in a little more detail some of the goals of
		the work envisaged.

  principles.txt: A few design principles and justifications applicable to
		the whole project.

  structure.txt:  Describes the component layers necessary to
		achieve the initial objectives.

  hierarchy.txt:  Describes some of the implications of having
		hierarchical clusters.

Then we have documents describing individual components in some detail:

  api.txt:	General requirements for the cluster APIs, including
		at least some details of inter-process communication 
		between cluster service processes on a single node.

  recovery.txt: A manager for the local cluster processes on a node.
		This must deal with both the initial startup of
		processes, and the coordinated restart after a cluster
		transition. 

  communications.txt: The cluster communications layer: getting nodes to
		talk to each other.

  integration.txt: The cluster integration layer: binding nodes into a
		coherent cluster.

  discovery.txt: The discovery algorithm used for cluster reforming in
		the integration layer.

  barrier.txt:	The barrier API used for cluster-wide synchronisation of
		arbitrary services.

  quorum.txt:	Quorum management: how to tell if it is safe to access
		shared cluster data.


NEW.txt: Stuff still to be integrated into the above