See also: IRC log
<DanC_EDI> presumed in transit: RF, DO
VQ: We are expecting visitors from U of Edinburgh and Fujitsu to meet with us at 10:30 on grid.
Previous minutes are linked from the agenda
Ed had comments, reflected already.
Norm: Good useful minutes
RESOLVED approve minutes of last meeting, Tim abstaining
[discussion of agenda]
VQ: Goals of meeting are two:
Noah: Update about recent MS dev conference re web applications we could do 15 minutes on.
VQ: Next scheduled 5,6 Dec 2005, Cambridge, Massachusetts
Noah: IBM could help host if any problems.
VQ: The following one is in the Technical plenary week. We need to decide which days we should pick.
Tim: We have to decide in terms of other groups we want to meet with.
Norm, Henry: We can't manage this meeting, we have two other groups we are also committed to!
Norm: It has seemed helpful to sit down with other groups
Noah: Is it easy for us to be all there all week, maybe? At least those present (ex Roy, Dave)
VQ: We normally have incoming as well as outgoing members
Norm: We should remind candidates for election that there is already a scheduled meeting
VQ: How long should we meet for as a group?
HT: How about a session at the end to compare notes -- unless people have to leave on Fri pm?
Noah: Concern TAG not roaring ahead -- we need face time together.
Norm: Expect XML Core and XML Schema to be on one end and Query to be on the other end of the week.
[discussion of combinatorial logistics]
Norm: Propose half a day at the beginning and half at the end -- I could justify dropping other meetings for that.
Norm: Were a XML Processing Model WG to be formed, we might well want to meet with it.
RESOLVED We will meet on the afternoons of 27nd February and 3 March.
VQ: Now, how about the next meeting in May or June?
Norm: Possible offer for Western Mass
VQ: There is a international www conference in Edinburgh with an AC meeting before it in Edinburgh
<DanC_EDI> "18-20 May 2006, Edinburgh, Scotland, United Kingdom" -- http://www.w3.org/Member/Meeting/
<DanC_EDI> 22-26 May WWW2006 http://www2006.org/
VQ: Lets meet in June
Closest airports BDL Bradley in Hertford
Tim and Noah will drive of course - 2 hours.
RESOLVED: Hold those dates June 12-14 2005 modulo Norm getting a big enough tent etc.
Fallback is Kansas City 20,21,22 June
<DanC_EDI> (semantic web scheduling use-case: N tag members in a room, all reserving dates in N different calendar apps)
VQ: We should make a report.
Tim, Henry, DanC planning to be there
VQ, DO not sure
RESOLUTION: Henry will present the TAG to the AC meeting.
[Enter right Dave Berry, National E-Science Centre, Edinburgh, David Snelling, Fujitsu Laboratories of Europe.]
[Enter right David Orchard, dapper in a light suede jacket and fashionable dangling earphone]
[Dave S promises to email the slides to firstname.lastname@example.org] -- Slides
DS: Discussed with HT and DB and others about what to talk about. In "What is a resource, "Here be dragons" .
I have a few hats, and NextGrid are paying
this trip. I work for Fujitsu labs Europe
... Working as GGF vcie-chair
... Chair of WSRF TC in OASIS
... Was co-chair of OGSI working group before that
... I was with the Unicore project, a European parallel to the Globus project.
DS:I was wondering about using the word "resource" with different colors depending on the context and local meaning. I want to look at the object model for a resource and see what the overlap is with W3C, OGSA
DS: Slide: Outline.
I can talk about what a WSRF resource is. OGSA resource is a moving target.
Tim: "Modifyable by representation" only applies to information resources
[discussion of "information resource"]
DS: If two URIs differ, only the resource can tell you that two mean the same thing
Tim: well, you can find out in various ways
Tim "Representations don't have URIs"
[HST is somewhat surprised by that statement]
[more discussion ... how representations are just what is transferred in the message when the Information Resource (IR) is accessed]
DS: Generally the URI is insufficient to access a resource. You also need cookies etc.
[discussion of the notion of identity of a resource]
[discussion of why banks do or don't give bookmarkable URIs]
DS: In terms of the world
outside, the notion of context is very important
DS: Slide: Reality Check
Noah: If the provider tells you about the structrue of the URIs, then you can depend on that.
Tim: For example an HTML form permits the client to know some of the syntax of a set of URIs.
DS: Slide: WS-Resource
DB: A WS-Resource could be a terrabyte store, and the XML representatoin only some stfff about it.
HT: Like streaming radio ... the URL s for the description.
DS: A WS-R could be an executing
application. The state could have some of teh global variables
of the application.
... There is a notion of destruction of the WS-R which may or may not destry the resource itself, just the access to it.
... The semantics of destruction of a resource on the resource are not defined.
... A WS-R is a pair of a web service and a resource which it gives access to.
[missed about garbage collection of resources]
DS: Its type includes its web service type, but also includes the schema for the document
TBL: A general MXL document or a very stylized document?
DS: That changed last week - now it is general Xml document
DC: Why did it change
NM: What was the old way? I thought it was a clearly a set of properties, which allowed me to select a subset of properties. It used to be a property list model.
<DanC_EDI> (boy, if they had a property list model, I wonder if they considered RDF/XML)
DS: Yes, we had a notion of qname-named properties, though they may or may not have existed as a real XML file. It turned out historically, that initally the only way to access this was to query across all properties if the document was arbitrary XML. However, there has been a request to add a capability to put and get the whole resource property document. Now that we have that capability, then you can use it for an arbitrary document. Property lists are a special case.
The query now allows you to ask for
properties, and/or random XML. A query is given in a query
dialect (we only require and expect XPath support as a query
dialect) where the query dialect is defined by a URI
DS: I could have 2 bank accounts
behind the same service, and the End Point Reference may have
more info to distinguish the accounts.
... A common use, though, is for the extra information to be encoded in the URI, such as in the ?query string at the end of the path.
NM: We ahve been less happy with people using stuff outside the URI.
MN: I have not been happy with the discussions we have had with the WS-Addreessing
DS: I felt that it used to be that ref params are much softer, not really identification. Now we have [with the EPR] a single package and we know that all the ID is in there. But not all the EPR may be used for ID -- as a function of the EPR minter, the ref parameters may have things like quality of service, etc. The message as a whole is interpreted by the web service. At Fujitsu lab we use ref params to distinguish between different objects. I don't think we should. I think Globus does not - they put everything in the URI. I think IBM does dispatch on the ref params at the moment.
DS: Fujitsu's trouble with just using URIs was [missed]
Noah: Why would anyone want an EPR (except for acrrying metadta which we are or talking about) ? It was to be able to use qnames -- these are difficult to use in URI's as there snot space to put ful URIs
<DanC_EDI> ( dorchard, when you write something in our blog that's relevant to a TAG issue, I'd like you to mail www-archive a pointer with the issue name/number in the subject.)
<DanC_EDI> (I'm also interested in having some blog stuff make its way to www-archive automatically or something. but there are copyright issues)
<noah> what we want is something like:
TBL: There is an argument which says that the extensability of qnames was invented for languages like XML and RDF and query languages for them.
[discussion of opacity]
<noah> I think opacity is as seen by users...in this case, one of the reasons for this structure is for benefit of the resource owner.
DS: If using a URI to identify a resource, we are OK so long as the minter mainatins the notion of opacity ... why do I have to presve qnames in there? I can have a local hash table
DO: Yes but you are doing client-side queries
<noah> Resource owners may be able to dispatch handling and associate behavior with pieces of the URI, and the whole thing may still be largely opaque to the user, or at least as opaque as traditional short parm names.
<ht> HST records his understanding that the opacity property was applicable primarily to the unparameterised URI, not the bit after the ?, which seems by nature much less opaque. . .
<DanC_EDI> TBL: [something about WSDL and prefixes]. DO: yes, let's do that
<noah> Noah records that he thought opacity as seen by "user" is as much or as little as the resource owner says. I thought what we've said is that (a) in general URIs are opaque except insofar as pertinent specs talk about things like hierarchy and (b) even in cases where substructure is documented, you should try hard not to depend on it.
<Norm> Dave/TimBL think that declaring the prefixes associated with URIs in the WSDL would be a good idea
<dorchard> TimBL and Dave are interested in option #16, which has just the prefix in the URI.
<DanC_EDI> option #16 among what?
<dorchard> the blog that I posted on qnames/uris
<Norm> This allows them to be opaque to the user, but the WSDL can still inform minters how to do it
<DanC_EDI> "16. Query and non-lexical variant 4 with no ns decl" in http://www.pacificspirit.com/blog/2004/04/29/binding_qnames_to_uris
TBL: Maybe if you really feel you have to tell a client to exprerss the semantics of a bit of XML or RDF in the query part, then the WSDL could contain a set of default namespace prefixes which are expected in the client part of the URL.
DO: Yes. I definitely agree to that.
The guy building the web service knows what schemes he can understand so it's not a problem that this scheme doesn't allow arbitrary extensibility
Noah: The URIs are then not quite opaque in the way we discussed, because the resourec owner has allowed the client to understand it
<noah> Noah points out that WSDL should be positioned as only one of the ways of documenting the association between short names and URIs.
DS: I don't personally have a lot
of use cases where you have to go beyond the URI. There was (in
OGSA time) a persistent handle GSH unique for all time handle
with a resolver which went to a GSR. GSH have the property that
they are never reused. The GSR could be reused. That
architecture has been dropped. If you are using a URI as a
referener to the WS resource, then when you mint it if you used
HTTP... then the web service tooling which decides on the text
of a URI whcih can be accessed over the net.
For Example: http://fujits.com/atlum#abadeafa where atrim is the web service and the abadeafa
DS: yes. The EPR was designed to carry this. This worked pre- Ws-Addressing
<dorchard> I mentioned at the June f2f in cambridge about needing a uri query language, http://www.w3.org/2001/tag/2005/06/14-16-minutes.html
<DanC_EDI> (zooming up, mentally... we seem to be having an interesting discussion of some web services specs, but I expected to get an answer to "what is the Grid" and I don't recall getting one. hmm.)
DS: The question was ... couldyou guarantee that you would get the details of what is on the end, but now with EPR you do
<DanC_EDI> (for "OGSA grid services" google nominates http://www.globus.org/ogsa/ )
DS: You can also use a/ instead of # but then some people had a problem with the feeling that you ended up with a web service for each resource,
DO: Because of the way their software was written
<Norm> OGSA=Open Grid Services Architecture
<DanC_EDI> (is OGSA an organization on which there is consensus on, for example, when it started?)
DS: In EPR form this would look
... <Address>http://fuj.com/arun/</address> <disambiguator> abadeafa</>
<dorchard> actually, it is *roughly* <To>http://fujl.com/arun/</To><disambiguator isReferenceParameter="true">abadeafa</disambiguator>
DS: Slide: OGSA Specifications
<dorchard> Specifically, the <address> field in the EPR can be used in various ways, and isn't sent as a soap header named <address>
<DanC_EDI> DaveS: ... OGSA WG ...
<DanC_EDI> (taking a random walk over OGSA WG stuff... they seem to use gridforge, which allows group members to upload... ugh... .doc files)
<DanC_EDI> (interesting... Screen share service (glance.net) http://www-unix.gridforum.org/mail_archive/ogsa-wg/2004/06/msg00077.html )
DS: Slide: Toward OGSA resources
<DanC_EDI> (teleconference minutes in .pdf http://www-unix.gridforum.org/mail_archive/ogsa-wg/2005/09/msg00033.html ... hmm... must be diagrams and such... phpht. nope. just headings, paragraphs, and lists.)
DO: You get protocol independence from using WS-Adrressing ... you may want to use soap.jml: scheme.. froma hsitroical persopecive, if the soap to http binding whuich ould do thinsg with rfef params then they could have used it,
<DanC_EDI> (... "OGSI days"... did I tune out of a history discussion?)
DS: A service may be defined in an EPR in teh OGSA architecture which you can call if the main address breaks to get a new equiavalent address in the future.
TBL: Why not allocate a domain name for the service?
DO: Granularity too low
<DanC_EDI> (I thought it was also about who owns the "job" vs who owns the computer where the job is running)
<noah> Noah also suggested that there is a hosting relationship: the same application may one at one hosting domain for awhile, then move to another. Tying the URI's used in the app to DNS names owned by a given hosting site may be artificial.
<Vincent> Roy joins
DC: If when I set up a web servcie on a provider, I quoted my domain name, ad then all te names were allocated by the servcie in that space, then life woudl work better.
[discussion of single points of failure and names and addresses etc]
<DanC_EDI> hmm... as IETF liaison, I'm obliged to keep an eye out when people are using W3C specs to re-invent DNS. hmm.
Tim: People often deplore the "single point of failure" in a DNS to HTTP redirect but then to get something more reliable they implement a atscak of DNS lookups and HTTP and SOAP access
DS: Well, we felt DNS was unreliable, so we built a fallback lookup system
TBL: Which uses DNS?
DO: Sounds like the XRI example for reasons for wanting to move things around. We never responded to the XRI folkd wanting to move things about how they should do it in HTTP
DS: OGSA are looking to have unique names, such that two names will never be allocated to the same thing.
<DanC_EDI> (hmm.... owl:differentFrom expresses ?X != ?Y ... maybe OGSA naming stuff could use that?)
Noah: Why of that a characteristic of the name rather than the deployment of the name? For exmaple in noah-mendelsohn.org I coul dcontrol it so that this is the case.
Norm: This is a social problem
Dc: It could be solved by for example using money as a fine for reeuse.
DS: They are looking at the social aspects but there are many different naming schemes. For example, http:/./handle.net/ but they don't like that because you have to pay for it, They want persistent names for free
DS: Slide: Observations
Tim: Do you mean IR not resource? yes
<noah> Meeting: Minutes of Tag F2F Afternoon of 20 Sept. 2005
<noah> scribe: Noah Mendelsohn
<noah> scribenick: noah