OGSA Interim Meeting -- 14-16 February 2005 =========================================== * Participants Jay Unger (IBM) Steve Tuecke (Univa) [14] Jem Treadwell (HP) Ravi Subramaniam (Intel) [14,15] Andreas Savva (Fujitsu) Manuel Pereira (IBM) [15] Steven Newhouse (OMII) Takuya Mori (NEC, ANL) Mark Morgan (UVa) Sam Meder (ANL) [14,15] David Martin (IBM) [15] Tom Maguire (IBM) [14,15] Fred Maciel (Hitachi) Hiro Kishimoto (Fujitsu) Andrew Grimshaw (UVa) [15] Ian Foster (ANL) [14,15] Abdeslem Djaoui (RL) On Bridge: Dave Snelling (Fujitsu) [14] Frank Siebenlist (ANL) [14,16] Ted Anderson (IBM) [15] Minutes: Andreas Savva * Should OGSA use WSRF, WS-I+ or both? The "2 stacks are unavoidable" argument (WSRF and WS-I+), ref: UK e-Science OGSA meeting. Note: - WS-I+ Core: WS-I plus WS-Addressing, WS-RM, BPEL - WS-I++ Core: Also adds WS-Eventing There are other related specs that are not in public or standard space: ws-trasfer, ..., ws-management, etc. They correspond to wsrf, ws-n, wsdm. WS-I+ or WS-I++ are not good names. A comment that there are actually 3 (or more) positions/approaches; currently articulated as 2 stacks. UK e-Science concern: Who will be the eventual winner. And desire to avoid re-work later; motivated by OGSI experience) - The danger is that choosing WSRF, without sufficient explanation and reaching out, might make some (sizeable) part of the UK e-Science community turn to something else. - Adopting WS-I as lowest denominator is safest choice - But any functionality over and above that would probably end up re-inventing WSRF and WSN mechanisms in a (different) non-standard way - Proposal: to define OGSA profile at a higher (more abstract) level than either stack and then provide 'bindings' to both - In other words, identify what features should be in the profile and then do the mapping - These features have been discussed for the past 3 years; not diving in to the definition. - Already have described requirements in good enough detail in the OGSA version 1.0 document - This OGSA informational document might have to be adjusted to allow for multiple mappings since it currently states WSRF,WSN,WSDM, will be used. - Proposal: A lesser profile - That describes a smaller set of features (no wsrf (ws-resource), base notification, service groups, not constrainted to metadata) so it would probably just add something close to addressing, and ws-eventing. - The problem is that what is pushed out will still have to be described somewhere; - If it is left up to each system designer then no interoperability; not a solution. - Not enough experience to decide what is the best approach? So it is either choose an approach or wait. - There was an argument that this topic was discussed before (at great length) and settled already. - Rough consensus: - Go ahead with a profile and that profile should be based on WSRF because the main expertise (within the group) is there and because these are the specs that are in the standards process - If anyone wants to soften the language on WSRF in the main ogsa document (for version 2.0 or 1.x) and/or base profile: Open a tracker (OGSA v2 or Agenda Topic) with the *recommended revised text* - Action for Ravi, and Steven N - Arrange a tutorial/study session on how WSRF and (WS-Transfer, etc) fit together, how and where they overlap. ** Why WSRF or not (revisited on Day 2 during summary of Day 1) - Tony Hey feedback on previous day discussions and UK e-Science support for dual model. - Projects in UK already doing things using WS-I; e.g., for job submission, etc. - Unless OGSA-WG offers a more compelling reason for using WSRF these and other projects (e.g., EGEE) are probably going to use 'plain' WS - Need to understand objections to WSRF in more detail. - Andrew is following up on some of these issues by phone; but this is work for more than one person. - Key WSRF concepts (naming, state exchange, introspection...) seem fundamental. If WSRF or something like it is not adopted then different projects will probably come up with similar functionality eventually out of necessity. - Proposal: Service specification documents produced by OGSA related activities should be WSRF agnostic so as to allow for different implementations/mappings. - Proposal: Each party to provide a brief statement on why it is choosing to adopt WSRF * Basic Profile Document walkthrough Ref: https://forge.gridforum.org/projects/ogsa-wg/document/draft-ggf-ogsa-basic-profile/en/2 ** Overview by section - Agreed to keep the referenced versions as they are; and - Lay out a schedule and time frame and how to sync with final versions specs - Talk to WSRF and WSN chairs and find out their schedule. - (Presumably all namespaces would have to be consistent for the final version.) - Action: A timeframe document for the BP (Tom and Dave S) - WS-RMD - agreed to remove it completely No document available yet so it is premature to include. Do not be prescriptive on this feature. - Resource lifetime - agreed to keep the section - How far to mandate immediate resource termination; still an open issue and will have to talk about it again. - Difficulty of testing it but agreed that resource lifetime should be in. - Still need to discuss the details - If created due to a message exchange and so on... - ServiceGroup - agreed to drop the section completely - Base Notification - keep - Issue: base notification namespace (WS-Addressing namespace update and topic update) - Options: 1. re-do schemas for this profile 2. Ask WS-N to update their specification > Action for Tom - Base faults - agreed to keep - But probably just to mandate that the defined schema should be used - Security - Issue: Allow anonymous interactions - Issue: release date of BSP 1.0 ** Detailed by compliance statement - R1002: Issue: Check that the policy element (wsp) has been removed from the WS-Addressing schema and remove this statement. > Tom to verify. - R1003, R1004: may be removed pending ongoing work within WSRF TC. - R1005, R1006, R1007: - Should a reference be passed in (with the meaning of don't give me this reference again, and how far that should be mandated) and related issue of how to compare references. - Should resolver EPR also be part of the reference handed out. - Covers a large number of cases, but obviously not all. Agreed that it is useful to have. - Ideally this feature should be defined in a lower level specification (but no candidates yet) - Idea that if references are broken then the entity introducing the change should be made responsible to 'fix' things up (so that they can be re-resolved) - ogsa-rns:Resolve - Need followup discussion on how to factor/refactor the existing RNS spec to provide the naming resolution spec needed by OGSA. - Need to define namespaces (ogsa-*) for schema and wsdl - 3.1.6 to be deleted - 3.2 (MIH) - Probably what's in the WS-Addressing is ok - 3.3 - Faults - Faults with resolving resilient references should be also defined. (For now put in this profile since there is no candidate spec.) - RNS might cover part of this but probably not the faults specific to updating resilient references. - What are those faults? - Ideally there should be a new spec. for this space. - 4. Requirement for specific embodiment; but even WSRF TC likely to just use single WS-Addressing embodiment. - Proposal to delete the section completely and assume a specific embodiment (and state which earlier on) - Rejected for now. List the embodiments that will be used in this profile. - Issue: Require usage of WS-Addressing or not > Tom and Ravi - Implicitly the text seems to be assuming WS-Addressing anyway. - Ok, provided final W3C version is used eventually - 5. Resource Property - xpath 1.0 response issue (undefined) and whether this would be an issue. Need to determine what and if there is interoperability issues. (Might need to define the response in the profile.) - Agreed to look at using nodeset serialization in xml digital signature as base for this. - Topics - Value-change notification topic: (don't need to have except) - RPDN[?] - Add notifiable resource properties property - Set resource property - Action for Tom to describe what is implied by supporting this. - Immediate destruction (revisit later) - Revisit 8.1.1 at a later point - Security - Issue on whether to have a section here or to make it a separate profile. - Consensus that some basic functionality should be described here. - Agreed to extend section to cover more than session security. Also support other XML security mechanisms. - Agreed to support both message or transport; use either or both - Agreed to split this off as a basic security profile. [XXX] - WS-I copyright for front matter; too much text is being reused. Hiro and Tom to followup on copyright issues and redo if necessary. [Day 2] * Naming discussion RNS: started as file system specific. Generalized and reduced to cover naming (Human name to Abstract name; and possibly to Address (aka EPR)). Mainly the 'resolve' operation. ** WS-Naming: - A string (xml). Globally unique in space and time. - Define a way to be able to compare for equality; probably a byte-wise comparison (note: i18n support) - "Abstract name EPR": EPR that contains an "Abstract Name" as extensibility element (not in the reference properties since those are opaque) - Optionally contains 0 or more resolver EPR - Do not have to be used; alternative resolvers could be used - Not aiming for a new specification but a convention or a profile. - Resolve operation takes an EPR and returns an EPR (or list of); and constrained to return equivalent ones if an abstract name is included. - Optionally perhaps a set of convention on what resolvers are included (resolver of choice, last resort resolver). E.g., factory may be a resolver, etc. - Issue: Type, etc, information should or could also be included? - Proposal: Do this as a WS-Naming profile and refer to it from the OGSA Basic Profile. - So the relevant pieces now in the BP are split off. - It is worked on within the OGSA-WG; no need for a separate WG. - In combination with an RNS specification, which may have wider scope. The Profile uses and restricts the usage of the 'Resolve' operation Action: Andrew, Mark, Manuel to do a short write up. Review options to move forward including the possibility of doing something like a BoF on {WS-Naming, RNS/Resolver}. - (Might be able to use the existing slot rather than applying for a new BoF slot.) ** RNS update - Revision based on Dec F2F discussions - Definition of resource properties moved to appendix - Application of RNS to filesystems - Agreed this should be moved to a separate document; if RNS is referenced in a Naming profile it would look strange to reference a spec talking about filesystems. - This is also a good argument for spinning RNS out in a separate WG. * Data activities update - Update from UK Data F2F. - Full minutes by Mark are on Gridforge: https://forge.gridforum.org/projects/ogsa-wg/document/OGSA_Data_Interim_Minutes_2_Feb_2005/en/1 - Proposed BoF for 'ByteIO' : simple posix IO like file access - Operations: Read, write, append: they take an offset; but there is no session state with clients (no file pointer) - They are not the 'user' interface; these are the back-end services; hence no 'open' operation. - Aiming for a minimal set; extensible to cover more complex scenarios. - Planning for BoF at GGF13 * EMS ** EMS activities update - Minutes from UK EMS F2F: - Full minutes by Mark are on Gridforge https://forge.gridforum.org/projects/ogsa-wg/document/OGSA_EMS_Interim_Minutes_1_Feb_2005/en/1 https://forge.gridforum.org/projects/ogsa-wg/document/OGSA_EMS_Interim_Minutes_2_Feb_2005/en/1 - Basic EMS documents Background: It became obvious that OGSA version 1.0 EMS terminology had problems due to overloaded terms like 'job', 'task' and so on. There was an initial revision last year to change the terminology followed by an attempt to simplify in order to get a minimal set that could be quickly defined. UK meeting took things further by further reducing the already simplified text. The resulting draft (including some UML) was posted to the list during a previous call. http://www-unix.gridforum.org/mail_archive/ogsa-wg/2005/02/msg00009.html - A proposal for a "Basic Execution Service" interface was later produced by Steven N. [To be circulated more widely soon.] This is one input for the upcoming BoF. Other input also welcome. ** OGSA-BES (Basic Execution Service) - A Service interface for 'job factory' or container or .... It corresponds to one small piece in the EMS picture (service container). - Note that this is not a scheduler or queue (out of scope) - Job state/modelling is in scope - Review of early draft by Steven N.: - submitJob: - Input: jsdl 1.0 - Output: a job identifier/"Abstract Name" - Faults: (feature not supported) - and a variant - submitJob - Input: an abstract name pointing to a jobdescription - The abstract name could be an EPR (a WS-Address) - Why not state that it is an EPR? - Just trying not to be overly restrictive at this stage. - This is an open issue. (Need an alternative normative definition of WS-Address that is not based on EPRs if this is not to be defined as an EPR.) - getJobStatus - jobTerminate - This is an operation on the container not the 'job'; the job might also have a similar operation. - Shouldn't this be done through WS-Resource lifetime mechanism? - Or why create new operations or terms for functionality that can be provided by existing mechanisms? - Not intended as a substitute for a destroy on the created resource (job). It is an alternative (shorcut) way of managing jobs within a container. - Input could be a list of abstractNames; provide a serviceGroup like functionality. For example, to quickly purge all jobs from a container. - ... still open... 'different' names for what is already available/possible using EPR and WSRF model - getSupportedStatus - Provides a view that is more easily mappable to more traditional job submission/control functionality. - Simplest set that is useable by a large part of the community; and can be delivered within a reasonable time. - Requirement: exactly once job submit semantics - Does not have to be part of the spec, but the spec must allow support for it - Possibilities: A jobid (or some unique input token) could be in the submission document or part of the API - Differences in opinion - Concerns with possible overlap or clash in functionality with basic profile definition. - In the sense that the BP takes the WSRF route while this proposal might take a different approach that does not use WSRF, but provides similar functionality to what WS-Resource modeling would provide. - And also 'way of doing things' that fits with the OGSA/OGSA-EMS model/architecture. - But this is something that should and could be done as the work of this proposed WG progresses. - Whether the 'container' has operations that can manage jobs or whether that management is done directly on the jobs (and only that way is defined) - BoF followup - (Andrew is leading the activity for now) - Select chairs - How to make sure that this work remains OGSA compliant - Milestones: draft by GGF14; implementations by GGF15 and go into public comment; proposed recommendation by Dec. ** Basic Execution Profile - Two approaches: - list candidate specs currently available - what we would like to be able to define and then match what is available with what is needed to be able to describe it - Currently available or nearly available - JSDL - WS-Agreement - JSDL and WS-Agreement connection. A need for some simpler normative document that describes how these two fit together. - The OGSA-BES BoF/proposal - Proposal for a short informational document to describe what the space looks like, where the holes are and whether they can be filled. - In other words an EM roadmap document that will be part of the OGSA roadmap - With a basic execution profile following after that. - Which part of the broader EM architecture should be worked on next? - Refinement could proceed along with other activities to do a proto-profile - Proposal to begin BP-EM after June; after the OGSA-BES activity starts up. - Three activities: 1. OGSA-BES BoF spawns WG for BES Specification 2. OGSA-WG works on Basic execution profile (With input from existing specs (JSDL & WS-Agreement, etc)) 3. OGSA EM Design team works on Refactoring of EM architecture * GGF13 session review - Confirmed sessions and session owners - Outreach: - Overview: Hiro - Profile: Dave S - Data: Andrew, Dave B - Basic Profile: - Tom and Dave S - Naming (check possible conflict with CGS session; for cross-wg collaboration---not on naming but resource management) - Could use this as a 'BoF-like' session for WS-Naming/RNS - Try to arrange to do GFS-RNS in combination with this - EM : the 3 activities (Ravi, Andrew?) (Paper on resource properties: Andrew to write up?) - Deliverables for GGF13 --- initial draft submissions by 18 February - Roadmap draft for first session (Ian) - Draft Basic Profile (Tom, Dave S) - Draft Basic Profile Template (Dave S, Tom) - Draft for Naming (Requirements doc) (Mark ) - Draft Data document (Mark to see if he can get something together by Friday since Dave B is away) * Roadmap document - Look into combining the two existing documents (Original draft from GGF12 and the profile creation proposal) > Ian will do draft - Need schedule input from EM, Data, Basic Profile team leaders to update the draft before GGF13. [Day 3] * Design teams - Hiro's presentation http://www-unix.gridforum.org/mail_archive/ogsa-wg/2005/02/msg00073.html ** Resource Management - CIM or WS-CIM adoption as model for OGSA Main work must be done by the respective design teams (Data, EM) with help from the RM team. - Proposal: 1. Informational document comparing CIM and GLUE. 2. Informational document on how CIM could be used for OGSA work, which are the relevant pieces, how they should be used. 3. Use/refer to CIM in OGSA profiles - Additions, or proposals to CIM if gaps are identified could be done directly to DMTF - CIM the model in contrast to CIM renderings (WS-CIM, and so on). Could use the model or pieces of the model as appropriate without worrying too much about the availability or not of a good way of rendering it for use by WS. Action: Fred to lead the activity, send schedule to Hiro. ** Security - Security profile was split off from basic profile. - Security profile work: interest by other parties in GGF. - Frank to lead this activity. ** Information Services - There are related activities within GGF, e.g., the INFO-D WG. Hence the joint OGSA-INFO-D session to work out the relationship in more detail. - Most things needed to build on are probably already in the basic profile - Could start working on a profile for information around GGF14 * Teleconference schedule - Initial proposal by Hiro http://www-unix.gridforum.org/mail_archive/ogsa-wg/2005/02/msg00074.html - Reviewed and modified. - Action: Hiro to send out updated version - Action: Hiro to check that next week Monday is a holiday or not for US participants (President's day)