Posted by
Phil Henshaw-2 on
Mar 29, 2008; 4:02pm
URL: http://friam.383.s1.nabble.com/Wed-Blender-Stereo-and-Computational-Photography-Videography-for-Cultural-Preservation-tp525972p525986.html
Marcus,
I think the boundary conditions of the problem include both the variable of
system design and control, and that of the independent behaviors of the
users. The question is what each of those contributes. With computer
networks you can't do without both, of course, but you can consider what the
options are for each independent of the other. Then both may learn to make
a combined system work better.
You say " ...resources can be managed by a secure executive process that
divides up the work fairly. Systems that don't do this are non-critical
systems." That is generally true for computer networks. Playing God and
deciding what is fair is the practical thing in that circumstance, since the
system emerged as a controlled system to start with.
The network manager might be really 'out to lunch' some times though, and
the users needed to share the resource without that global view and central
control. What could they accomplish just between themselves, is the
question. They'd have virtually none of the information the manager uses
and none of the control. If left to themselves, how would they do it?
I think they'd develop usage signals of various kinds, that communicate
things like 'here I come' or 'now I'm done'. That would help optimize the
use of the resource without making the users talk to each other to figure
out and mesh each other's needs. There are signals that convey these kinds
of messages in natural systems, like usage growth and decay patterns, which
telegraph what will follow in considerable detail if you look into the
derivative rates. That forecasting ability then allows responses before
conflict arises. If users did that then everyone could get more out of the
shared resource without dropping chains from overloading the buffers, or
having to talk to each other.
To me there are several things here that are closely analogous to the
problem of optimizing the interaction between users of less well defined
resources. For independent users of resources in open environments there's
no 'God' person worth their salt as controller. Not knowing how to do
without that role seems to have become a primary problem for the economies
and all our complex shared uses of the earth in general.
Does that make sense?
Phil
> sy at synapse9.com wrote:
> > Now that since nearly anyone's taking of more resources is
> increasingly robbing and disrupting other users, has sort of become the
> main source of conflict on earth..
> >
> If someone wants to copy a real big amount of stuff from one node of a
> cluster to another (there can be tens or thousands of these nodes), the
> switch can connect these two nodes. All other transfers in the system
> can be going on without notice of this. To the extent other people
> want to deal with those two nodes, the switch can fairly divide down the
> bandwidth between those people. This will typically be a small
> fraction of the total capacity of the system or the network.
> Furthermore, on a typical large cluster, there will be a parallel
> filesystem with many independent block devices and very low latency
> switches. If I have 100 nodes all writing at once to 100 different
> block devices and there is a effectively a different wire from the node
> to the drive, then there is no contention.
>
> If a hundred users all want to do this, with their respective
> entitlements, and from different nodes, then at some point you run out
> of gas. But a hundred users rarely if ever all want to do this.
> This is a pretty standard assumption of many kinds of telecommunication
> systems.
>
> So, neither wire networking nor bus use is usable for your analogy.
> The reason is that these resources can be managed by a secure executive
> process that divides up the work fairly. Systems that don't do this
> are non-critical systems.
[ph]
>
> Marcus
>
>
> ============================================================
> FRIAM Applied Complexity Group listserv
> Meets Fridays 9a-11:30 at cafe at St. John's College
> lectures, archives, unsubscribe, maps at
http://www.friam.org