DAIS Telcon Minutes 19/02/04 ---------------------------- Minutes by: Mario Antonioletti (mario@epcc.ed.ac.uk). Attendees: Allen Luniewski (IBM) Brian Collins (IBM) Mario Antonioletti (EPCC) Dave Pearson (Oracle) Inderpal Narang (IBM) Vijay Dialani (University of Southampton) Norman Paton (University of Manchester) Sastry Mallady (Oracle) Agenda ====== Points arising from minutes of the Redwood face-2-face meeting Progress on activities for GGF10 Sematics and Metadata Spec documents DAIS requested session for GGF10 on WS Resource proposal Scope of data management in DAIS DAIS roadmap beyond GGF10 --- o Points arising from minutes of the Redwood face-2-face meeting Minutes distributed were distributed by Susan. Dave: included in the agenda in case anyone wanted to ask any questions on this. ... ok so there are no issues. o Progress on activities for GGF10 - Sematics and Metadata Norman: on the semantics front nothing in particular has happened within the last week. Previously some interaction between Simon, Norman and Mario. Have been trying to get the key notions. Have a small collection of core concepts (terms) that have now been included in the core spec. This only produced a little extra text. Other than that not too much extra progress. The next step will be to nail the semantics as to how these terms inter-relate and that will require more work. Hope to have made more progress describing the relationships for GGF10. Dave: seem to remember that something would be published this week. Norman: don't remember that this... Mario: Dave you gave Simon the option as to whether to publish something this week or to leave it until the slides were more polished off. You wanted to prevent too much unnecessary work in case the content required iteration but Simon preferred to spend a little more time thinking and polishing the slides off. Norman: Simon was concerned that the slides by themselves would not be that helpful. Hope that the text would help with this. ... My feeling that the way to proceed is that we will work over the fortnight leading up to GGF10 trying to polish the slides for GGF10 and present the work there. ... As with Metadata ... I interpret this to refer to the DMTF collaboration ... Susan has updated the scenarios document and sent it through to Larry to see if he wants to put it on their list. Don't know if he is going to ciruclate it on their list. We could circulate it on our list but would require to get comments very quickly... Dave: I have some comments ... have been talking to a chap from DMTF and it took them 3 weeks to define a database ... not reached the resource granularity... Norman: they have schema on their todo list... the CGS group have offered to pick up and run with some of the DAIS requirements. They will act as the interface with DMTF.... one part of the document that Susan is distributing deals with deployment and the other with discovery. Have restricted deployment to SQL matters ... Dave: we have two options - we publish the document or CGS does ... Norman: Susan has seen it that CGS are quite keen to own the document ... we could publish it on both lists and let the CGS people run with it..... Dave: Have tried to come up with some text defining the line between management and DDL. There's an issue on how you deal with strings. Brian: Patrique and I have had some discussions and outlined another line of management that deals with quality of service ... there is a class of managment where you are telling the system what to do and not how you do it while there is another class of commands that deal with how you do a task which affect the QoS. That line of reasoning may help ... thinking of updating stuff that has been written with Patrique ... what do you think? -- Sastry fell off the telcon at this point. Dave: DB administration starts where you start dealing with the overall resources and the architecture ... and also the QoS .. most DB managers have the privileges to create schema, update tables and that class of operations I see as being valid in DAIS but the problem is that you can cut in several ways ... our approach in the specs has been fairly minimalist ... tried to produce some wording which was independent of the relational DBMSs. Brian: in terms of the scenarios you are right ... maybe I'll continue having discussion with those that are interested ... Mario: is it worth posting to the list and open up the discussion? Brian: will talk to to Patrique and see if it's worth adding some words on QoS ... and then open this up and put it out to the list. Dave: what has been proposed could be put up as management.... Vijay: not sure how the service model that we are proposing with the level of detail we are looking at works ... are we looking at a closed or an open system ... if we are providing these services how does one define the autonomy? Dave: not in scope to define dabatabse architectural constructs ... that would be part of management ... what the services should provide ... DBMSs are already fairly sophisticatd that already have associated privileges with them ... do these things go in an administrator model ... there is a deficiency that there should be a portType to manage the underlying infrastrucutre .. as well as the service Inderpal: we need to nail down the data managment portyType... ... Vijay: agree with the use case but if we give different portTypes different security model then we will only be able to find these things at run time ... the model requires a bit of refineement. Dave: the problem is that we pass everything as a string in the current model ... have discussed this with simon. If you are passing through strings then it's up to the implementation as to whether they do any string checking ... Mario: I agree with that ... Vijay: I beg to differ ... we have adopted a stop gap ... assume local queries will be executed in a particular way ... I think things will evolve into a more sophisticatd way Dave: do you have any suggestions? Vijay: I do not have a complete model but I believe that the model has been evolving ... Dave: if we don't use string passing then we would have an explosion in portTypes... Vijay: I don;'t think the solution is to adopt a parser ... Dave: but everyone agrees that that is an implementation decision ... that's an implementation decision ... Norman: it's quite a difficult position which we are exploring ... if we look at the two types of management: managing the service or the underlying resource ... what we can build on from what is being produced by the DMTF manages the resource ... if we seek the DMTF stuff which is all about descriptiona as opposed to action then that will provide us with a working model but then we need to sort out the service properties but then this leads to an inconsistency about the granularity of how we describe things and the granularity of what we can do with it ... when you ask of what one can do to manipulate these things you end up being able to do only a little with the operations that we have ... so that you could argue that we are being inconsistent .... can get consistency by leaning on the SQL specs ... it's a complicated space ... we are not trying to import huge amounts of work on to the DAIS group so that's why we are leaning on DMTF. Inderpal: can we build adaptors to manage the data resource ... we should focus on the data service management .... Dave: it would be easier to manage that the data service but as Malcolm pointed out at the ANL f2f that you also have to manage the underlying data resource somehow. Concern that DMTF is an object model and we may have problems getting consistency ... Allen: uncomfortable about pushing too hard in trying to manage the underlying data resource ... there are lots of difficulties ... Dave: by management we are talking about a class of operations that can be considered as management - privileges about the objects, the creation of those objects and little more than that ... that would be in the scope of DAIS everything else should be out of scope for DAIS ... the resources that are relevant to us are relevant to those described by CGS. Allen: I see those things that are coming out of the data service... Inderpal: index would not be exposed through a data service ... QoS is handle autonomously by the data resources .... Dave: what if you are doing a distributed query and we want to create an index to have more efficient process ... Vijay: there are various forms of distributed query processing ... there is a shared state that cannot be made visible in a Grid environment Norman: depending on the level of service you may require more/less knowledge about the data resource ... Inderpal's suggestion is that yoou provide enough information to phrase a request ... if you want then to find out how much is going to cost then you require another level of information ... now you could do the 2nd step by exposing more info about the data ... which is what Vijay is arguing against or you could do it in an operational way where a data resource would take in a query and return the relevant type of information. Within a Grid environment you would want to support this amount of information. Dave: I don't think that it's just an issue for distributed queries might want to perform a scan of atable as supposed to a read ... if you constrain the nature of the service then in a commercial environment you would not use a Grid as most of what we are talking about is already available in a DBMS... Inderpal: but these things are pushed down by a data service.. that's my current thing ... I don't think we should be exposing scan vs index they should just provide quality of service ... Dave: I dont think that these things should be exposed to the user but the application may need to know certain characteristics about these things ... if you publish 20 services for accessing the same data an application will require QoS data to determine which service it will use Inderpal: you really require a higher level description ... Brian: can expand and open my previous email and expose to the list ... would like some input from the XML side as well... not aimed at producing something for GGF10 Dave: so if you evolve a doc that describes the discussion up to date - how DAIS should deal with DDL? ... Brian: not for GGF10 ... not enough time ... will mail to the mailing list and for the next GGF will have something more concrete to push out. Norman: so do we include Dave's material on to the doc? Consensus is that it will be incorporated. - Spec documents Mario: have made some changes to all three docs. I then passed the docs on to Norman came up with some further issues, as below: - Adding examples of the terms as extensibility parameters to the createService activity. - Nothing much is said about notifications. Perhaps we simply state in Section 3.4 that notifications can be associated with all SDEs in the specification? - Not done anything to management, in the light of the ongoing discussions - I am hoping that someone who has been at the relevant telcon can implement some changes. - In the XML specification, not always sure of the relationship between the boolean result and some of the faults. Thus I can't tell when an operation may return with false to indicate that a request was not completed, and when it might return with a Fault of the 'any other fault' variety. - The relational and XML specifications are not consistent in terms of how they descibe parameters to createService operations. I've not had time to address these since then these other than adding Dave's management text but I think that there are no show stoppers there. Amy said that she could provide the terms as extensibility parameters to the createSerivce operations but she wondered whehter this was worthwhile doing if it was all just going to be stripped at the next iteration of the spec. I passed on the relational spec to Susan as she wanted to add stuff about stored procedures. I can pass the core spec to Amy and I guess she can have a look at the XML issues. It was agree that: Susan would get the relational token. Amy would get tokens for the core and xml specs. Norman will interact with Amy regarding any modifications that need to be made.. - DAIS requested session for GGF10 on WS Resource proposal Sastry fell off the call so this issue was not addressed. o Scope of data management in DAIS This item was addressed earlier on in the call. o DAIS roadmap beyond GGF10 Dave: you brought this item up so maybe you could say a couple of things. Mario: there is a concern within OGSA-DAI that the DAIS specs are not converging and that they may not do so within the lifetime of the OGSA-DAI project. Even if they do then we will require some time to be able to implement them. DAIS could very much benefit from the feedback that this would provide. If the specs do not converge then there OGSA-DAI may disinvest some effort in the standardisation procedure as there is no direct benefit in trying to do this... Norman: what was stated at the last f2f is that we will seek to reorganize the specs to minimise the dependency on all other specs by GGF11 and we should also be able to provide some information as to how these things map down on to things like WSRF and other frameworks ... the operations and metadata will be there and be stable ... the specs just now, as written, are slightly more abstract then the WSDL so this should be possible .... Dave: currently ogsa-dai aligned with GGF7. Have funds within OGSA-DAI that deal with the standardisation effort ... Mario: so it is fair to say tha by GGf11 OGSA-DAI could start safely implementing the DAIS spec. What if the OGSA Data Service document were to be radically factored by then ... how would this impact on the DAIS specs? Norman: we have some control over the ogsa data service documents... there should be an updated document for GGF10... there will come a time when the ogsa data service will stabalize .. should be stable ... the DAIS specs will then be revised to take into account the ogsa data services spec .. by then no one will have the right to radically modify this document ... I don't think that anything is going to happen to the data service spec. ... Allen: the WSRF does not change the data services spec in any particular way ... Norman: our intent is to get the core types nailed by GGF11 .... and that whoever wants to implement it will be in a position to do so by then. Dave: OGSA-DAI can influence the spec ... Norman: OGSA-DAI should not feel incredibly defensive ... it's got a lot of influence over the specs. AOB --- Norman: (to Inderpal) there was talk at the last f2f about trying to put some effort to describe the scenarios Inderpal: talking about 3rd party delivery? Norman: no Inderpal: related to data movement? Dave: you said that the scenarios should be exposed to a wider community... Greg's work with simon was going to act as the basis for this and Maha was going to formalise it ... Norman: is it going to make as a draft for GGF10 by next Tuesday? Inderpal: yes Norman: the mechanism for up-loading these documents is that they get sent to me on Tuesday and then I upload. I hope that you, Allen, are able to do that for the data services document ... Allen: I'm not hopeful Norman: that will be an issue .... Dave: when do you want the documents Norman? Norman: by 6pm uk time would be a good thing ... Dave: I will send an email Norman: Allen we have a session in GGF10 on this document and we will need a document ... Mario: will not be here next week so someone else will have to take the minutes. Dave: Norman and I are staying in the maritime* hotel for GGF10 .... Inderpal: was the roadmap discussed? Dave: that's what Mario brough up earlier on in connection with OGSA-DAI.