Minutes of the Data Design team telcon, 27th October 2004

0) Summary of actions
- Manuel to post new version of GFS spec by end October (if possible).
- UVa team to read the GFS specs. 
- People to register for gfs-wg mailing list for notifications.
- GFS team to review UVa documents.   Manual to be point of contact. 
- Dave to send reminder and pointer to naming requirements documents
- Dave to send deatils of December OGSA F2F meeting
- Dave to schedule discussion of data movement for a later telcon
- Other assignees to produce use-case scenarios for use of metadata.
- Andrew, Allen and Dave to continue with a strawman architecture.


1) Early discussion

Present
Andrew Grimshaw (U. Virginia)
Mark Morgan (U. Virginia)
Shannon Hastings (OSU)
Stephen Langella (OSU)
Leo Luan (IBM)
Ted Anderson (IBM)
Manuel Pereira (IBM)
Dave Berry (NeSC, note taker)
Osama Tatebe (AIST)
Susan Malaika (IBM)
Allen Luniewski (IBM)

Apologies 
Simon Laws (IBM)

Minutes approval: https://forge.gridforum.org/projects/ogsa-wg/document/Minutes_6th_October_2004/en/2

Dave added an item to the agenda to mention the OGSA F2F planned for December.  Discussion of the metadata scenarios was postponed.
        
2) Update on outstanding actions
- Dave to arrange telcon that includes GFS-WG & EGEE.  [No success arranging an extra telcon but GFS people are present today]
- Everybody to review the naming requirements documents  [Dave to send reminder and pointer]
- U. Virginia team to put their implementation on the web [Done]
- Dave to chase Brian and Malcolm re data movement [Done, and first draft written with input from the EU NextGrid project. Dave to schedule discussion for a future telcon.  N.b. Brian is retiring from IBM]
- Mark to produce a use-case scenario for EMS use of metadata [Done]
- Other assignees to produce use-case scenarios for use of metadata. [To be continued]
- Andrew, Allen and Dave to continue with a strawman architecture.  [To be continued]

3) BasicFiles and Contexts
- Discussion in the light of the U. Virginia implementation

- Everybody is familiar with the document.
- Mark summarised the implementation.  Interfaces to services kept simple.  Put a lot of logic into client-side library. Used WSRF.NET (as familiar to UVa).
- File interface: read, write, append, truncAppend.
- Factory portType for creation
- Client-side library with "usual" commands (ls, cd, cat, cp).

Q. Relationship with GFS-WG spec?
Perception on the OGSA side that GFS scheme is file-specific.
GFS scheme can actually handle other junction types (e.g. databases)
UVa Contexts do have a notion of type in the directory (e.g. for directory listing and for error checking).  More or less irrelevant for server side.
GFS junctions are general: with subtypes for physical junctions or logical junctions.

Key architectural issue: junction points and entry into file systems.  Logical junctions are abstract names.  When dealing with files, the unit of export is often a file system.  So what happens when you transition from one scheme to another?  E.g. does the OGSA scheme overlay the underlying file system name space?  GFS doesn't (so far).

UVa approach: units of interest to applications are files.  Therefore need to name them.  The fact that files are in file systems is orthogonal.  Some implementations will have other resources that are containers (i.e. file systems, directories).  From a modelling point of view, the paths in the UVa path need not match the underlying file system (although this is an obvious implementation approach).

GFS spec in two parts - one general, the other specifying the types that are useful for file systems.  It seems possible that the UVa approach matches the GFS generic level.

WSDM have defined file system types; may be worth the GFS people looking at this.
GFS work has been done with links to NFS v4 design.

Architectural issue: Does the fact that an element is a junction has to be exposed to the client?  Is this relevant to the client?  One junction type is a pointer to another context service.

GFS scheme differentiates between directories and everything else.  Possibly a difference with the UVa scheme.

We agreed that each group needs to read the other's specs/code in more detail, vearing in mind the issues raised above.  It may be that we are in violent agreement.

Actions:
- Manuel to post new version of GFS spec this week (if possible) because it demonstrates abstract points more clearly.
- UVa team to read the GFS specs.  
- People to register for gfs-wg mailing list for notifications.
- GFS team to review UVa documents.   Manual to be point of contact. 


4) OGSA F2F 
 8-10 Dec in Washington DC.  Discussion of all OGSA (i.e. not a DesignTeam F2F).  Will include interaction between data services and execution management services.
 Dave to mail details


5) Wrap up (5 min)
      AOB
      DONM