NorduGrid technical meeting

18, 20 January 2005, Tallinn

Minutes

Present: Balázs, Anders, Marko, Ilja, Aleksandr, Oxana, Juha, Mike, Arto, Farid, Mattias (at the minutes).


1. Meetings

a) Next Meeting

Next meeting will be co-located with the NGN meeting in Vilnius
(around 20 March, precise date TBD).

b) This meeting's programme - Order of discussions

Indexing services and storage systems on Friday. Rest on Wednesday.

c) Decide on a time and place and participants for the *dedicated*
   data management workshop

Data Management workshop proposed to be at CERN beginning of March.
Joint meeting with some EGEE data management people. Need to prepare
input for the workshop in advance.

d) Rome meeting: what shall we propose, represent

Rome standardization meeting 17-18 February. Common jobsubmission
interoperability. Balázs and Aleksandr (probably) will go.

e) EU Grid conference: who, what

EU grid conference, Amsterdam. Need to compose a programme for the
NorduGrid sessions (2 times 90 min). Balázs is responsible for the
program (should be submitted by Tuesday). Presentations form users
(projects) of ARC middleware. Mattias will present SweGrid, Arto (or
someone else from Finland) will talk about M-grid. The Danes have
their own session in the conference (DCSC/DCGC).

f) Other: Globus World, Taiwan, etc

GlobusWorld 7-11 February: Aleksandr
Taiwan: Oxana and Farid
ARC VRVS meetings Tuesdays at 14:00 CET (15:00 EET)

2. Current commitments (personally)

Everyone confirmed their current commitments...

3. Software

a) Status

Information System: Documentation. Configuration and start-up scripts.
Longterm plan: Replace GIIS with native LDAP.

User Interface: New version based on arclib under development. New
version has file locking and wildcards. Benchmarks implemented but not
tested.

Monitor: Caching and sorting implemented in the monitor. Not yet in CVS.

Logger-interface: Work in progress.

Gridftp-server: GACL for grid-jobs - documented in XRSL manual. Still
not possible to define GACL for output files - this need some
rethinking on how output files are defined in the xrsl. Suggestions
are welcome.

SSE: Some memory leaks. Ready for testing.

Logger: MySQL server slow. Better after memory upgrade. But table
redesign might be useful.

jarclib: Minimal functionality. Testcases work. Will be used by java
GUI.

RTE repository: Need to be filled with information about existing
RTEs.

b) NorGrid comments.

Hello World type jobs too slow - fixed in 0.5. We should set up more
test sites with the 0.5 version where users can test the
functionality.

Interaction mechanism - uploading and downloading files to/from
session directory possible. If more specific use cases are defined we
are willing to help.

Portals vs. cli - "The ARC cli is simple and powerful..."

Inexperience of users - We can offer to organize i tutorial in Norway.

ARC easy to break by small changes in configuration - ask for more
detailed bug reports/feature requests in bugzilla. We could/should
create a test script/utility that checks configuration consistency.

Enabling running services non-root, requirements:

- slapd: file permissions in control directory
         LRMS access to information for all users
- grid-manager: submit job as another user
                read/write to session dierectory
- gridftpd: permission in control/session directory
- nordugridmap: write access /etc/grid-security/gridmap
- crlupdate: write access /etc/grid-security/certificates/

c) Release milestones, todo list, - roadmap

Release 0.4.5 will be done soon.

Need massive testing of 0.5.x before tagging 0.6.

To be done for 0.6:

- Globus 3.2.1 (or 3.9.4). Globus 3.x needed for opteron processors.
- Infosys startup (remove globus-config)
- Job states
- New CVS
- Monitor
- Configuration examples
- Changelog
- Release notes
- Migration documentation

Build platforms:

- FC 1-4
- RH 7.3, 9
- RHEL 3 (+ x86_64)
- SuSE 8.2, 9
- debian 3.0, 3.1
- Mdk 10.1

d) Makefiles for the new CVS, how to

Shall we use the full GNU chain?

aclocal
autoheader
automake
autoconf
libtool
make

We would like to have libtool for indirect dependencies. Some autoconf
features would make life easy for differences in e.g. pthread
implementations.

Anders will make a proposal for the build machinery.

4 & 5. Indexing Services and Storage Systems

What do we have:

- storage servers: gridftp, https
- index servers: RC, RLS

gLite: Replica Meta Catalog (but not documented)

Aleksandr's document describing the "Floating Objects" service was
discussed:

Single distributed integrated storage and index service.

Each service contines objects that store location of data or links to
other objects to form groups. There are also reference objects
refering to objects on other servers.

There should be a protocol defined for communicating between object.

Servers can be made responsible for creating additional replicas if
some replicas becomes unaccessible. Replicas can be tagged with
expiration times so that if the time expires all replicas except 1 (or
some other numeber) are removed. Possible to implement some
"hysteresis" in this case.

It should not be a hierachical system - to avoid single points of
failure. (But it is possible to build one using this system.)

The objects are responsible for initiating copying and deleting of
files. No files should be actively removed by clients.

Partially implemented by Aleksandr (https + soap).

6. Article for publication

Should be discussed on the mailing list...