xml-dev - RE: Microsoft's DISCO proposal and XML packaging

["Bullard, Claude L (Len)" <clbullar@ingr.com>]

Hi Simon:

I read it, and thanks.  Interesting.  I think it 
is a good start and eminently recognizable which 
is why I trust it.  If it were truly revolutionary 
with no basis for comparison, I would reject it 
as too far ahead of the event horizon for immediate 
use.  As tis, I know what to do with this: engineer 
an enterprise for business.  

This is the enterprise engineering stuff.  Hooray, 
XML finally gets away from the period of domination 
by parserHeads and into the really truly fun bits 
for application engineers.  The wait is over.  We 
can finally boogie in the big arenas.

To the WaaaaayBack machine, Sherman.

The seminal period for me in thinking about this was 
1988 to 1992.  If you could poke around the CALS and 
PDES archives, you might find quite a bit of the thinking 
from that period.  At that time, we were looking at 
non linear dynamic systems and exploring their realization 
in real time systems for describing very large integrated 
product development environments, particularly, how 
hypermedia could be used to enable these.   The problem 
to be solved was not just cost reduction and high 
quality, but the issues of noisy environments.  How 
to hold a band together on a stage and still be 
able to improvise at will is the same problem as 
managing a business through a period of technical 
emergence:  predictability depends on discovery 
of known good sources of information.  Gotta learn 
how and deep knowledge isn't always as good as 
awareness, desperation, and dumb luck.  

But it should be.

One example at the time was enabling high-tech plants in 
the countries just then coming from behind the crumbling 
Soviet bloc where 50 years of Marxism had created 
low resolution work environments (non-competitive 
and very difficult to change or retrain). In the face of a 
dysfunctional culture made that way by accidents of 
history, how does one shape the behavior 
toward a stable cooperating system?  Marxist/socialist 
systems did precisely the opposite.  They tended 
toward obsfuscation to hide mistakes instead of 
translating mistakes into learning.  This is also the 
"to the metal" problem of teaching XML by the way 
sometimes just called, NIH.

Competition is essential to local coherence 
in communications.  They have to WANT to work.  A whip 
just won't do it.  The pigeon bites the hand 
that holds the whip AND the food after a while.

There are some fundamental concepts:

o  Event-based.  Signals are typed and are point to 
point.  This enables the system to not require locked 
synchronization but does enable scheduling.  Locked 
synchonization depends on centralization and that 
impedes local discovery.  Take a spontaneous ride on your 
instrument while playing Beethoven's Fifth and 
see who is first violin next week.

o  Rules and Contexts:  the orchestration model is 
pretty directly the model the hytimers discussed.  
Remember that was a music description language 
to begin with and the understanding that the 
timing and gestural model of music could be generalized 
to an orchestrated performance was seminal.  Take a 
great ride during a performance of Don't Get Around 
Much Anymore and unless you step on the guy next to you, 
you can be the first saxophonist next week unless 
he does a better one this weekend.  Context and rules 
count in a negotiated set of services and roles. 

(I didn't see a role model in SCL.  Hmm.)

o  Hierarchical description of business processes as 
contracts.  This is simply a Work Breakdown Structure 
with discoverability.  It enables a percolation model 
for performance.
 
When percolating, you can't predict an exact path 
but you don't have to.  Virtual time is top down and 
real time is bottom up.  These are view dimensions.  
The upper level views are managerial/control views 
and the lower level views are real processes.  The 
idea is to enable the scheduling of opening and closing 
views that have tests for well-performed behaviors, 
aka, goodness or the reliability or trust in the 
information.  See Claude Shannon:  "Data becomes 
information as it removes uncertainty" and Boltzman's 
equation, S=KlogW in which the number of good referents 
in the system determines its entropic state.

o  Binding points.  We talked a lot in those days about  
tesselating models.  It is a geometric concept but it 
explored the idea of DTDs (now schemas) that enabled 
point to point constructions.  The idea is that the 
schemas could be bound as needed when needed but more 
important, adapted within a defined space.  Schemas 
define boundaries.  We think of these now as namespaces 
but the concept is the same: non-ambiguity in an address.

Timing plays a big role here as well as loose and 
tight coupling such that orchestration is not overly 
constrained and the resulting performance is not 
overly predictable (Is Entertaining thus keeps 
the Attention (the real commodity of cost) of 
those who must perform it and those that choose 
to attend it)).  

Compare classical music performance 
to jazz or rock performance to get a feel for why this 
has to work this way.   Latency and noisy environments 
are very important.  The question is, how did Ringo 
keep time for the rest of the band in the face of 
a few thousand screaming girls?   He watched John 
and Paul's buns.  They wiggled as they played so while 
he could not hear, he could keep beat.  This is a
gestural system lightly coupled to ensure a reasonable 
if not perfect rendering.  It was a discoverable 
service of John and Paul's buns which Ringo could 
use when all else failed due to noise.  Ringo was 
the timekeeper and he scaled time to the motion 
of the buns to ensure real time coherence (on the 
other hand, not great musicality if you listen to 
the Hollywood Bowl performance but the customer 
could care less, and folks, that is quality).

If we use the notion of the time quanta, it sets the lowest 
level of process resolution and is scalable.  The timeline 
can be said to exhibit the features of a Cantor set, or 
in a two dimensional realization, a Sierpinski gasket.  
Mappings onto the coordinate space of events that fail are not 
addressable, or simply, fall through the cracks, or better, 
are not viewed or modeled.  Closure of a process creates a 
continuous map.  Failure to close is a discontinuity.  
It is important to remember that a hierarchy of infinities 
is not a real object; it is a recursive process or a nesting 
of recursive processes: loop to success or exhaustion. 
Transfinite numbers are neither real nor really numbers, 
just a way to talk about accuracy or granularity.

Human systems can be viewed a loosely coupled real time 
systems where policy directs events.   Time is a not a 
director, just a scaling frame.  Events direct events.  
If time to reply is not important, rough granularity is 
acceptable (again, synchronization of very large systems 
is problematic due to latency).  The document based 
systems or policies create a rough closed feedback process loop 
for adapting the communications to do it or do it again 
until it closes.   Latency is the key issue not just 
in speed of send/respond, but also in absorbtion of the 
signal in presence of noise (roughly, the power law 
at the receiver).  Self-correcting systems (eg, 
shared schema) are a means to detect and correct 
for noise.  Self-adjusting systems (eg, dynamic 
schema) enable the system to self-correct or to 
learn and thus evolve new capability.  In genetic terms, 
self selection.  Note, that in Darwinian thinking, 
only living systems have local rules for self-directed 
evolution.  It is a fascinating idea.

The problem of the web is superstition, that is, the 
web is an amplifier that feeds signals back into itself 
and this enables degraded modes to emerge and be sustained.  
Little differences become give differences and events 
occuring in non-visible dimensions create produce 
effects that amplify across the boundaries.  That is 
why I get so wrapped around the history thread from 
time to time because in Darwinian thinking, the history 
of events is accidental but affective.  Not understanding 
this leads to more superstitions and more incorrect 
behaviors.  The system may still be sustainable but 
its direction is questionable.  Feeding error back 
into a system is stochastic composition with a 
pseudo-deterministic model of known processes. 

Information does not want to be free (superstition); 
Information wants to cohere (a prediction based on a 
range of sustainable communications).

Think of it as a fractal event stream (feedback-mediated, 
but false values in the control range).  Cool for 
some compositions, but not exactly what the price-sensitive, 
weWantWhatWeSpecified, results business models are 
supposed to predict and as a control, produce.  In 
other words, if we want a high goodness factor in 
the nested business model, we need trusted patterns 
and trust is a markovian function.  We can introduce 
an episodic model depending on how much error the 
processes can tolerate and still close correctly. 
Episodic models do not have to close with absolute 
precision as long as 'the job gets done',

To define the processes of the enterprise (discoverable 
services), create a bounding defintion or mission 
statement of goals, then hierarchies of processes 
to meet the goals (as defined in that document 
element in the contract).  Remember, this is not 
a simulation model although one could create one, 
it is a computable living contract for real 
business processes.  It alerts the humans if it 
detects something amiss (opportunity for negotiation 
and learning or more discovery).  Drift in the 
model means costs, so renegotiation is always an 
option and why I asked you about the negotiation 
model in SCL.  Again, virtual time (project time 
and project costs) is top down and real time 
(performance time and performance costs) are 
bottom up.  Renegotiation also entails taking 
a mal-behaving process and putting it to sleep 
so another task can use the resource (a foldable 
procedure).  If the process is dependant and 
a circularity develops, the performance locks. 
Min/max local states must be detectable (instrumented).

If we adopt the notion that these processes as 
declared are geometric, than they are space filling 
and the process space is bounded (can be said 
to have an energy budget), thus the notions of 
binding points, discontinuities, and so forth. 
The performance will have a distinct shape, but 
it should be of a class of shapes for like 
performances.  That provides a visualization of 
the goodness of the performance instance.  It 
may also point to the kind of math that can 
be used to analyze cluster density based on 
the links and link types.  This is the fractal 
stuff;  look there, not simply at 
storage behaviors because it might let one
predict network saturation.  If instead of 
just plotting points, one thinks of a 
Koch curve as a copy, rotate and scale 
operation, the process fills the space of 
addressable points in a space of variations.

The goal or Product is the classification 
characteristic.  Product and process wind 
together, inseparable but separated, like 
a double helix in DNA.  A query can be seen as 
a digital enzyme for state maintenance.

That's enough for today.  We may want to 
come back, dump the geometry, and talk 
about the relationships between style 
systems and orchestration.  Or maybe 
someone else wants to pick this up at 
the head and riff awhile.  :-)

Len Bullard
Intergraph Public Safety
clbullar@ingr.com
http://fly.hiwaay.net/~cbullard/lensongs.ram

Ekam sat.h, Vipraah bahudhaa vadanti.
Daamyata. Datta. Dayadhvam.h


-----Original Message-----
From: Simon St.Laurent [mailto:simonstl@simonstl.com]

I've not had the chance to read it in detail, but at least it seems like an
interesting start.