Re-thinking "Discovery"
|
|
|
David Booth
W3C Fellow / Hewlett-Packard
1/23/2003 |
|
Motivation
|
|
|
|
Existing diagrams are ambiguous |
|
Vague or conflicting definitions |
|
E.g.,"Service Provider" and
"discovery" |
|
Triangle diagram is misleading |
|
"Publish" and "Find" strongly imply
directionality. (All that matters is
that the two parties agree on the WSD - not who sent it.) |
|
"Find" is not always needed. (Often "GET" is enough.) |
|
|
|
Motivation (cont.)
|
|
|
|
Too much emphasis on "discovery"
role |
|
Not needed in many scenarios |
|
People think "Discovery Role" means
UDDI. |
|
Current draft and model ignores
semantics, even though semantics are present and relevant |
|
Explicit acknowledgement of semantics
also helps clarify where choreography fits in |
|
(Choreography language allows more
semantics to be machine processable) |
|
What I Did
|
|
|
|
Carefully analyzed 5 general
scenarios |
|
(3 main + 2 variations) |
|
Identified actors, actions and
artifacts |
|
Defined terms |
|
Drew diagrams that accurately model
them |
|
Described sequence of operation |
|
Conversed at length with several WG
members to get feedback, clarify, explain, etc. |
|
Scenarios & Diagrams
|
|
|
(See
http://www.w3.org/2002/ws/arch/2/10/roles_clean.htm ) |
|
|
|
FAQ 1 [Re: Scenario 1]
Isn't there still "discovery", even if it happened earlier
(static)?
|
|
|
No.
"Discovery" refers to the transfer of information about the
WS. |
|
"Discovery" of WS should not be confused with human exchange of
non-WS info. |
|
I.e., using the term "discovery" for
other purposes confuses the issue.
E.g., humans exchanging phone numbers. |
|
|
|
FAQ 2 [Re: Scenario 1]
Isn't there still "discovery" even if it is no-op in this
case?
|
|
|
The scenario models only the actual
roles that are required by the scenario. |
|
It does not model null roles that are
not required by the scenario. |
|
Any null role ("discovery", "foo",
"fum", whatever) can be hypothesized and added to any model, but
doing so proves nothing. |
|
|
|
FAQ 3
Can't these diagrams be combined into a single architecture
diagram?
|
|
|
|
My opinion: |
|
I don't think so |
|
Not without being misleading or
inaccurate |
|
Except for a VERY simple diagram with
only two actors (client and service) |
|