Matthew Dillon dillon at
Tue Mar 14 20:18:01 PST 2006

:from time to time I try to imagine how it would be when DragonFly's 
:clustering became reality.
:Is the goal to be able to just stick in a new DragonFly box and then the 
:power of the "cluster machine" increases? I mean, I wonder if there 
:would be any configuration requirements for new machines in the cluster.
:Or would it be enough to just stick a network card into a random 
:machine, compile with "options CLUSTER", connect to the network and 
:voila? :)

    The goal is to be able to boot up a random DragonFly box and assign
    cpu, memory, and disk resources to a named 'cluster'.  The resources
    then become part of the cluster and would communicate with the cluster
    through one or more connections (e.g. through a TCP connection or
    something more sophisticated).  Then, with that in place, we want
    maintainance functions to allow us to take the resources offline,
    put them back online, make them bigger or smaller, and so on and so

    Processes running on a DragonFly box would be assigned to one or more
    clusters supported by the box, or would simply be native to the box
    itself (that is, not be assigned to any cluster and operate on the
    standalone portion of the box).

    A machine would be able to supply resources to multiple clusters 

    The basic idea is to make the 'resources' as simple as possible, and
    thus as secure as possible relative to other things the machine might
    be doing.  We will start out with 'memory', 'cpu', and 'disk space'
    (abstractly represented by a block device or a file or whatever).  There
    will be an abstraction to support basic devices such as a network 
    interface, a pty, abstracted block devices which are made up of resources
    supplied by contributors to the cluster, and so forth.

					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>

More information about the Users mailing list