Notes on GFS-WG/RG meeting at GGF9 on Oct. 7 2003 Date: 10/07 Time: 6-7:30 Note Taker: Pradeep Padala Legend ------ Q - Question A - Answer C - Comment J - Jane O - Osamu Ar - Arun Leo - Leo Luan Andrew - Andrew Grimshaw Reagan - Reagan Moore Ann - Ann Chervanak Brian - Brian from IBM not sure of the last name P - Pradeep Agenda ------ *) Introduction, and discussion about charter *) Survey of Grid File Systems by Pradeep Padala (UFL) - SRB by Reagan Moore (SDSC) - Grid Datafarm by Osamu Tatebe (AIST) - Distributed StorageTank by Leo Luan (IBM) - SlashGrid by Andrew Mcnab (Univ Manchester) *) Grid Content Management by James Myers (PNNL) *) Virtual Filesystem Directory Service by Leo Luan (IBM) *) POSIX-like OGSA/SOAP Services by Arun Jagatheesan (SDSC) Meeting ------- Main presentation - Intro to the WG/RG by Osamu, Jane, Arun Slides ... - Pros/cons of WG, RG i) start RG spawn WGs ii) start WG - RG charter - WG charter Q: Describe what is application specific meta data A: Profile file system usage of applications and use the metadata C: Infer characteristics of applications from profiling applications. Working groups choose specs rather than technologies J: O: Metadata is related to hierarchical name space Ar: Users can describe/associate meta data with logical names. J: Way to describe metadata Q: Andrew Grimshaw: Logical name spaces for other things like storage Leo: name space for different data with same file system semantics C: Andrew: legion, avaki do this. In earlier OGSA discussions, file system service and other Reagan: In SRB, file, data base objects etc. can be registered J: Collect useful stuff from research projects like SRB and commercial things Andrew: C: Is it nomenclature problem? files vs content vs directory services Andrew: Don't want to constrain to only files David Martin: Many groups are working on directory services. RG vs WG An RG can study different directory services Q: What is reliable POSIX ?? J: O: It is necessary? To make it fault tolerant C: POSIX is semantics. Q: Will the FSS be part of grid services? J: We have heterogeneous data sources. Virtual directory service virtualizes and interfaces to various data sources and Grid FSS can be built on top of it. Also, traditional interfaces like NFS can be provided. Q: Are the data sources defined similar to what DAIS did J: Yes. Focus of the group is VDS and POSIX like interface Arun: Q: When would one use DAIS service vs GridFS posix service? DAIS is planning to define similar posix service. Leo: File sys is heirarchical C: In DAIS discussion being done for a front-end to http, ftp and it has hierarchy. Leo: Is the scope too broad? C: May be DAIS shouldn't touch it. But it should be consistent Leo: DAIS is talking about access C: Global naming is out of scope for DAIS may be J: It provides POSIX operations on global name space Reagan: There is a different between POSIX operations on logical name space and POSIX operations on storage resource C: file system is like dbmgmt. Files are like tables. Interfaces probably should be defined by same group. Reagan: In SRB, interfaces are seperately implemented Good work being done out side GGF - WebDAV, AFS, DFS etc. Andrew:Why can't survey be done first C:DAIS-WG, OGSA-WG might be a good home for some of the things David Martin: good to do RG for surveying different implementations Arun: Some of the process is already started. We already have a survey document. Some projects already did good research in requirements. David Martin: Adapt from some commercial or already available implementations instead of starting from scratch Votes ----- For WG: 8 For RG: 23 Ann: Scoping issues with DAIS or OREP C: Dais currently doesn't have file systems in charter. GGF needs to think about this. Andrew: different groups shouldn't be doing the same thing Brian: I don't know which way to vote. What is POSIX like API? Architecture that clearly shows what this POSIX API does? Reagan: Some implementations already have it. Survey document might answer this. Specific Presentations ---------------------- Questions during survey Leo: How to make sure that the input from the projects is correct? P: Actually there is no easy way of doing this Reagan: Are you suggesting that pradeep setup a test bed? Andrew: Instead of asking the projects to send data, why can't you survey based on papers? P: That's possible. There are too many projects to survey. Often, the papers don't have enough details.