DAIS Session 1 - GGF 11 - 07/06/04 ================================== Chair: Dave Pearson. Note taker: Mario Antonioletti Dave welcomes people to the DAIS session. Agenda for this session 1. Introduction. Review of Activities to Date: Dave Pearson 2. Grid Data Services Specification Update: Norman Paton 3. Discussion of Grid Data Service Specification Issues: Norman Paton ---- 1. Introduction. Review of Activities to Date: Dave Pearson A copy of the presentation is available at: http://forge.gridforum.org/projects/dais-wg/document/GGF11-DAIS1-Intro/en/1 Dave Pearson presents an overview of the scope and goals of DAIS. OGSA-DAI is mentioned as an eventual reference implementation of DAIS. A timeline of DAIS activities through previous GGF meetings was presented. For GGF9 the DAIS documents were refactored into separate documents: a core document, a relational and XML realisations that extend the core document. Dave publicises the fact that a files BOF is taking place at this GGF which may add another realisation. Dave also publicise the Infod working group meeting. Other things of note: - Minor changes have been made to the DAIS charter since GGF10. - The other DAIS sessions scheduled for GGF11 are advertised. - The documents produced for GGF11 are mentioned - an updated core document - an updated relational realisation document - an updated xml realisation document - a document discussing different mappings of DAIS to different WS specs Dave introduces Norman Paton who is to give an update on the changes that have been made to the core specification since GGF10. ---- 2. Grid Data Services Specification Update: Norman Paton A copy of the presentation is available at: http://forge.gridforum.org/projects/dais-wg/document/GGF11-DAIS1-CoreAndIssues/en/1 No one is expected to implement the core specification directly - this is done through the realisations. The name of the specification has changed from "The Grid Data Service Specification" to the "Web Services Data Access and Integration". This reflects the fact that we see the specification as being relevant outside the Grid Service area. There are implicit and explicit references to other specifications. Sastry Malladi: does it currently have an explicit dependence on WSRF? It does in the mapping section. The group however does not have an official position on WSRF yet. This mapping is thus currently indicative. DAIS has been working with other working groups - e.g. the DMTF through the CIM GGF WG to prevent replication of work. Q: when it says that it is focused on structured data what about APIs that expose data resources as structured data - does this specs deal with that? If there is another type of data resource then the specs have been factored in such a way that a new document can be produced to cater for this. Q: what determines the realisation - the data resource type or the data format? i.e. a relational resource could expose data as xml ... This would be an implementation detail. The documents are currently all drafts but hope to make these specifications. The specs are informed by the "OGSA Data Services" and "Scenarios for Mapping DAIS Concepts" documents. The latter two document will remain as informational documents. Since GGF10 have removed the dependency on OGSI and have reorganise the documents to try and remove external dependencies. A data service is a web service that corresponds to one of the DAIS specifications. Q: does a data mining system count as a data resource? i.e. it is a sink ... you have defined a data resource as being kind of broad. It could be. If a data mining system exposes a suitable API you could make it into a data resource. Maybe it should be tightened up. Norman goes over terminology (specified in the slides). The specification does not mandate how interfaces are composed into services. You could put relational and xml interfaces on data resources that support both models. Guy Rixon: in the context of a data service is there an implication if you revisit a service that the underlying database is the same, i.e the names of things do not change? The spec does not state that the names of things do not change. Malcolm Atkinson: this is an important point. Identifying the data is important Need to determine whether you are talking to the same resource. Need a specific global naming scheme - if you use an end point resource as in WS-Addressing then there is no requirement that the EPR will point the same thing. Norman: the first part of the spec is mapping independent Sastry: how you name a data resource should be independent of the mapping. Currently a name is exposed which is not the same as a WSRF end point reference. If you do not use WSRF you could a context token to identify the data resource. ... Maybe should not say anything about identity in this part of the presentation as this is tied to the mapping used. Tricky to resolve this issue without considering the mapping. The behavioural properties should not change over time as the semantics of the service are not well defined over this transition point. Do not have many data descriptions items as we are looking for something that is generic for all types of the data resources. If you thin that something has been missed out then do let us know. Sastry: how are the properties described in the WSDL? That is mapping specific. This currently is a resource property but it could be done in another way if you were to use another mapping. Behavioural properties are listed which characterise how the service behaves when interacting with a consumer. Norman again asks whether the list is complete and whether the definitions provided are adequate. Transaction behavioural properties are presented as an illustration of the use of behavioural properties. Sastry: WS-DAI - the core does not have a WSDL specification? Is there any WSDL? There is a little. If we were using WSDL 2.0 there might be some more. There is a lot more in the realisations. For the behavioural properties there are type definitions in the core spec that can be used. Guy: you do not seem to have any general interfaces for the curation of data that associate metadata with data? Not yet. Malcolm: ? (did not catch what Malcolm said but it was a supporting statement about what Guy had made) David Martin: that is the metadata question. (Question directed at Steve Langella who has had BOFs at two previous GGF meetings in order to set up a GGF RG.) Steve Langella: trying to form a group at GGF or possibly work within the context of the Semantic Grid... trying to resolve this for this GGF. Looking for a home at this point to carry this work out. David: it's a tough problem to even define what metadata models are. We have been associating our metadata with the data. Sastry: is the scope of DAIS set on accessing data resources or is it about how to virtualise the data resource in the context of the Grid? If it's the former what is the difference with existing products ... you do not cover the virtualisation... We are not trying to do something new as such ... Sastry: virtualisation is different ... you are trying to define an interface ... virtualisation hides how the data access is done ... Dave Pearson: this comes down to whether we are just putting a wrapper round JDBC and what the value of that is ... the owner of the data will determine what you can see ... in a scientific data environment you will want to give people access to these data resource and give people the ability to run queries ... this came out of requirements that came out of initial work in the UK ... Sastry: defining an interface to a data resource has nothing to do with Grid... We are going to get into a debate on the definition of Grids ... if we can characterise the question more precisely we can come back to it during the data services session. In the core document message patterns are presented that provide templates that are substantiated by the realisations. A relational example of a message exchange is given. Data factory allows representations of data to be exposed at the service end through a suitable interface. There are no operations defined in the core spec - this is left to the realisations but a set of message patterns are established the details of which are filled in by the realisation documents. We do not define data management at the moment. This arose from the OGSA Data Services document. There is work going on elsewhere. There are three aspects of management: - managing the resource - managing the service - managing the service-resource relationship David: do you think WSDM is going to produce enough detail? They are going to provide the framework. Sastry: WSDM are going to provide the framework - it is up to the working groups to provide the details. Brian Collins: there is a complex relationship of working groups ... we liaise with CIM which feeds into WSDM in OASIS... other groups are addressing this ... they may not do so to the required detail ... we need to keep an eye on what is going in that space .... Sastry: that is the right approach ... this group should not define management... Currently we have a holding position to see what happens in this space. Dave: currently out of scope but we would like someone else to put it in their scope ... Malcolm: this group should still define the requirements so that when a framework comes out we can see if this meets the requirements ... This is a sensible suggestion. We have been actively interacting with other groups. Some of the use cases identified by this group have been used to drive the work in the CIM group. Have avoided mapping issues as there is going to be a discussion at a later DAIS sessions. Only one slide is given showing aspects of the present mapping to WSRF. Malcolm: surprised that a data resource point maps to an End Point Reference... the web service may manage many resources ... you may require more than just the end point reference ... each web service may identify the resource differently and web services may use different identifiers ... we should ensure that if we have two identities they refer to different resources ... we could have the same name mapping to different resource through different web services ... In this mapping users see resources identified by end point references, an additional name property is available as an identifier... there is an issue then on how WSRF identifies the resource and how the resource is actually identified. More discussion on this is expected at the mapping session. Norman goes over the remaining DAIS session and sessions that are of relevance to DAIS. ---- 3. Discussion of Grid Data Service Specification Issues: Norman Paton The same set of slides were used for this part of the presentation. Norman lists some of the DAIS issues already identified: - have we missed anything out - are there anything that are should be addressed... ... David: WS-Security .... Sastry: WS-Security only gives the way of passing the stuff ... need an operation to pass the credentials.... Is this out of scope? Guy: if you want to use WS-Security you have to do per port type.... David: that is not a WS-Security issue it is an implementation... ... Sastry: you need an operation to get the credentials.... Need to look at the WS-Security and how this impacts on DAIS. Malcolm: there is an issue as to whether files are going to be incorporated within the DAIS framework ... There is a BOF to address this ... Malcolm: so it's still undecided ... This will be examined at the BOF. They will examine something that is going to fit in this framework. Still not clear whether this will function within this group or as its own group. Guy: given that your resource is a writable should there be behavioural characteristics defining who can write to it? We expose functionality that is provided by an underlying resource ... for relational resources if you have access to the database and are authenticated to write you will have the associated privileges ... these things are described by the CIM model ... you raise an interesting question as to whether this should be a behavioural property.... Malcolm: need an accounting model ... may have a 1Gb for 1 day or 0.1 Gb for 10 days.... the accounting model is much more subtle if the data resource is doing this ... if it's doing it then we need access to this ... There are lots of properties that would come through the CIM model that are related to management ... we may not provide mechanism. Guy: this is metadata relating to a resource ... need to have that metadata available at the interface That is a fair point. CIM does not have these properties available at the moment.... Malcolm: ... need to provide an interface to what can provide that kind of information ... need a way of enquiring ... ----