OGSA Data Services Teleconference -- 16 November 2004 ===================================================== Attendees --------- Dave Berry (NeSC) Andrew Grimshaw (UVa) Mario Antonioletti (EPCC) Ted Anderson (IBM) Leo Luan (IBM) Manual Pereira (IBM) Fred Maciel Shannon Hastings (OSU) Stephen Langella (OSU) Scott Loster Simon Laws (IBM) Mark Morgan (UVa, note taker) * Reorder Agenda to start on comments for FS work rather then meta-data work. * Metadata scenario due by Dec. 15 * Strawman Arch by Nov. 30 (something ready before Dec. F2F). * Dave in workshop in Netherlands on 30th. * Not aware of other work being done on strawman arch. Basic Files and Contexts ------------------------ - Last time, two groups (UVa, and GFS) needed to look at each other's work. - Initial impression seemed like a lot of agreement, but deeper delve needed. - GFS group about to produce new version of document (next week) - GFS group has looked at UVa implementation a little, but not quite ready to talk about. - Naming maps human readable to abstract names (AN). Another design team is working on the AN to RH mapping. UVa's prototype software is for the human readable to AN mapping, not AN to RH. -- Unfortunately, there isn't an AN right now so UVa's software actually maps human readaable to EPR. This caused some confusion last time. - What are the properties of an AN? - Uniqueness - Authentication - Aliasing - How do ANs get mapped? - AN could be a URN, a URI, etc. Some strings with lots of embedded stuff. - GFS has 3 level naming arch built in. -- VFDS designed with three leayers to exist simultaneously. -- You will have a logical name that could be resolved to middle layer, and then resolve that to endpoint, or you can go directly to endpoint. -- Question. Normally if you have a middle tier, you don't map directly from human readable to lowest because you would loose ability to migrate. - Even though the handles are scoped within a virtual organization, that does not mean that they aren't globally unique. GFS group doesn't restrict that. -- Impl dependent. - There are two implementations of the GFS stuff. -- One designed at IBM after the API. More compliant now with WSRF. -- A second impl by Osamu Tatubi's team. - UVa group should be able to get access to those impls for study. - New document will include the AbstractJuntion -- a more generic junction - OID parameter has been removed. WSRF now used to get statefulness. - Mapping of logical name to abstract name or address built into arch. Summary on Basic Files ---- - Seems like quite a bit of overlap. - Will details be the issue Action Items ------------ * Mark to look at next version of GFS document. * Mark to start revising draft document for UVa implementation. * Manuel to see if Mark can get ahold of draft implementations of GFS MetaData -------- - Last week, went through use case slides - Questions: Is a common ontology required? Is it a pre-requisite? -- No. Question is, how do I find services that are writeable, etc. -- Many ways to do this. You could set a writeable flag, etc. -- Question is, how do people understand this. - Have a common, agreed upon block of meta data that describes how this works. - Isn't a common ontology -- rather, a common starting ontology. A common base. - You agree on basic metadata and you begin from there. - Require this common root ontology or many roots. Should be atleast one - Service publisher should say that he is writeable and what he can write. - Next problem is saying, "I accept writes of an A, but not from you." -- Security involved. - Question: What language do you use to represent metadata. Ian Foster suggested RDF. -- Should be a common language. Suppose we use RDF and writeable means several things to several people * I could ask for a service that is writeable and get services that write A's, servies that cache it in multiple places in the grid, etc. - Service providers will have to agree on some base way to publish what their services can do. - Not saying that OGSA needs to publish a huge ontology. Only that there needs to be a root nugget to start with. - A number of different types of data exist with numerous ways of describing that data (XML schemas, database schemas, etc.) -- In these, you have to choose a single meta data representation. -- We need to choose at least one for metadata. - There is a WSRF metadata spec. -- Deadline is unknown -- proposed by Tom McGuire -- not in standards body. -- WSDM using pieces of it already -- WSRF folks will have a lot more info. - How important are semantics of metadata. Syntax is commonly important, but what about semantics? -- Depends on the domain. For service discovery it may not be important. For data type querying, it coudld be extremely important. -- In the context of a given community, metadata will have common semantics. -- Should have common ontology. -- People haven't given this a lot of thought yet, but think it's important. - Not aware of spec for replication metadata. Esp. for duplicating databases. -- Really shakey ground. Actions going forward ---------------------- * People review scenarios for metadata to find errors and mistakes * Simon to look at commonalities. -- Simon, describe general pattern and look at metadata people have suggested. * Dave to push on Architecture work if he can. Next week --------- Many can't make a meeting next week, mut probably enough can.