DAIS Session 2: Grid Data Services Specification and Management --------------------------------------------------------------- GGF10, 11th March 2004 Chair: Dave Pearson Notes by Mario Antonioletti (mario@epcc.ed.ac.uk) +----------------------------------------------------------- Presentations available from: https://forge.gridforum.org/projects/dais-wg Documents->GGF Documents/GGF10/Presentations Make sure that the document filter is working correctly. Unattributed statements in general belong to the speaker at that point in time. +----------------------------------------------------------- Dave Pearson introduces the second session. [See the DAIS-WG_Session_Update on Grid Forge.] Passes round the sign up sheet and makes people aware of the GGF IP statement. Two presentations. Dave Pearson gave some background information and set the scope of DAIS. Mentioned Information Distribution who are having a BOF as it was seen to be out of scope of the work of DAIS. Since GGF9 have been: - document restructuring - focus on semantics - DAIS requirements for CGS-WG - resolving non-mapping issues. Mapping of DAIS to resources which was the matter of the previous DAIS presentation. Information about the DAIS sessions still to take place was given. Documents produced for GGF10 - CIM Data Model for DAIS - DAIS usage scenarios - Information Dissemination in the Grid Environment - Grid Data Service Specification - XML/Relational Realisation - OGSA Data Service Documents May have a face-to-face meeting after this GGF. ----- Simon Laws on the Grid Data Service Specification. [See the presentation: Grid_Data_Services_Specification_Update on Grid Forge.] Talk about the specifications and then how we move over to GGF11. Examine how we describe the behaviour about our data services and an issue section. Talking about the specification ... basing our work on the OGSA Data Service document. Different types of model about how you might interact with a data service. Have behaviour that allows the server to hold a data set to which operations can be submitted subsequently. The current DAIS specs: - base specification - relational realisation - doing things with relational databases - xml realisation - doing things with xml There was not that much difference between the ggf9 and the ggf10 specs. The specs still use OGSI and the GGF9 OGSA Data Service interface partition. What we have changed: - removed references to WS-Agreement - have started talking about terms Separating form from function - using a refrigerator analogy the function is to keep things cold and the form is the shape, colour, etc. i.e. we do not want to worry too much about how the other specs we rely on are changing in our specs. We are worrying about these things but in a parallel track. We want to document functions that we wish to implement and later map these to an infrastructure. The other orthogonal thing is separating the base spec from the realisation specs - we have some base functionality and then some more specific types of functionality. It also gives us a place holder for new types of realisations which we/others amy want to add in the future. We will try to avoid fragmentation and incompatabilities. Active topic relevant to the base: - mapping to grid infrastructure - ogsa data services - data management - data movement - behaviour and controlling term (this session) - metadata - composite requests - transactions - security Pointed out that there is going to be a BOF about metadata later on in the week and there was a BOF on transactions earlier on in the day. A few words about behaviour and control. We initially worried about interfaces and not so much about their behaviour. Now we have been exploring behaviour. We started working about terms from WS-Agreement and WS-Policy and started playing with that list. This is not a complete list. We want to refine this and possibly other new ones. We start off with access where a client interacts with a service. Terms that might be associated with this type of interaction (listed in one of the slide). - AccessMode may only have Read/Write/Read-Write, Sequential/Random Access Some more descriptions of the terms described in the slides. ... It is not obvious which terms should be defaulted, which should be interacted with. We can also look at a Factory Inderpal Narang: the format of the resource? Could that be a term? Simon: yes. ... So terms that could be applied to a factory: - DataMaterialisation - Lifecycle - Relationships there are also subtleties. Inderpal: are partial result supported by DBMS manufacturers? Dave Pearson: some do ... ... Simon goes on to describe the factory terms (as outlined in the slides). Inderpal: is that related to provenance? Simon: I had not thought about it in provenance but it could be .... Description terms - sensitive/insensitive Notificaion - granularity - sensirivity but this may lie in the ground of information dissemination. We have some questions to answer how these should appear in the spec. Issues for the base spec: - do we have the functions correct? - do we have the metadata/description correct? - how do we describe expected behaviour? - what terms? - do we specify precisely in the spec and if so how? ... For GGF11 Need to: - decide what to do with the design principles in the document - achieve consistency with ogsa data servicess. - verify function scope - verify metadata/description scope - document behaviour model - model mapping to emerging grid infrastructurem, e.g. wsrf .... We welcome any assistance in crafting these specs ready for GGF11. Inderpal: making progress in the terms ... am I missing something ... Simon: yes ... we are making progress on all of these things but have just not described them in this presentation. Ammar Benabdelkader: Is there a realisation on object data? No I come from Amsterdam ... it would be nice to have a primitive. Dave: we welcome input from other groups. Norman: we have separate documents so if someone wanted to add another realisation we would welcome it. Ammar: My colleagues are heavily involved in workflow... Norman: we have an implementation of the spec for object databases ... Dave: we would welcome terms.... ---- Brian Collins gives a presentation about data management. [See the DAIS_and_Data_Management presentation on Grid Forge.] At the previous f2f when we were discussing the xml spec which initiated a discussion about management. The topic was not resolved then but a note was made that we should revisit it. A document was written for the CIM model which had some interesting use cases. Discussion continued in the telcons, mailings and a summary was posted to the DAIS mailing list. The current Grid Data Service Spec states that operations are passed into operations as string and then permissions dictate whether these operations are acted on or not. This, however, could mean that data management operations could sneak in. So what is the scope of DAIS? - should it include data management? - we need to agree what constitues data management? - ... other points in the slide ... Example scenarios for DAIS? (see the slides) Have been in the relational world for about 30 years now ... re-read some relational database texts for inspiration and came up with a list (outlined in the list/slides). Have tried to categorise these (see the slides) DAIS is all about specifying the what and the where. This would work the DML. Some of the DML statements also act on a logical schema. In the CIM paper the term data administration is used and I think that might be a better term ... there is confusion about data management as to whether it is addressing the service or underlying resource. There are other ddl statements that act on the physical schema. I would put the dividing line between management and access these two. This was put out to the mailing list. Inderpal: what if there are multiple physical schemas and one logical schema? As in a case federation ... access is to the logical schema but there could be multiple physical schemas ... I would like you to take this into account. Patrick Dantressangle: but this would be lower down. Dave Pearson: but you could use terms to define what type of operations you can perform. Inderpal: but you should take this into account... Norman: why is Inderpal making the distinction between the local and non-local ... ... Dave: you could be using a single logical schema that could be hiding multiple heterogenous physical schema .... Brian: Should the mode of operation be consistent across data types? files? XML? ... Finally which are the groups/specs that are affected and where should data management reside. Should this be handled by another working group? David Martin: there is CCDLM. Susan Malaika: we have had dealing with them - they got in touch with us at the previous DAIS f2f ... they are more configuration and deployment rather than management. Brian: sugges that we use the mailing list to engender discussion. There may be a small document possibly aimed at GGF11 or this work could be wrapped back into the core spec. Could keep it separate or fold it in ... encourage participation. Dave: useful for Brian to present this. We keep on identifying bits and pieces that are not in the scpe of DAIS ... we should be setting the requirements that other groups can pick up. Work this through in the mailing list. ... Any other comments? None were offered and the session was drawn to close.