NorduGrid technical meeting

October 1-2, 2002, Oslo

Present: M.E., A.K., B.K., O.S., A.W.

Minutes by O.S.

 Distribution
==============

support tar-balls and RPMs (CVS checkout, bootstrap, 
make dist -> tar.gz, rpm -ta)

client RPMs: self-contained, includes necessary Globus client parts

configuration: problematic

In general: good to have Globus inside, at least for binary distribution
for those who don't have any

Very important: provide ONE SELF-CONTAINED CLIENT, containing minimal
necessary Globus libraries (a la Netscape with Java) 

 Name : "standalone" or "lightweight"?

Apart of that: server RPM, client RPM, monitor RPM - assume Globus is there. 
Tarballs will be there, too.

Configuration: README must have a detailed description.
Templates, public certificates: Nordugrid-specific

 Tagging
=========

autotools -> 0.3.6 (NOW)

Next:

 - Redhat 6.2, 7.2
 - Globus 2.2
 - ngmove

All this will be tagged as 1.0 (mid-October ? Anyway before the Linkoping 
tutorial)

 MDS 
=====

anonymous query only

we stay so far with MDS

everybody uses the latest MDS version as soon as Balazs OKs (2.2.2) - by Monday

Glue schema: immature, but developing towards NorduGrid. It may work, but
difficult to assess now.

 Parallel jobs
===============

many dfferent clusters, different local policies and solutions. Impossible
to accomodate everybody; we may go for one definite example (Grendel?). Will
try (who?)

 Interactive jobs
==================

two kinds: 

a) users interacting with the job, e.g., via a GUI
b) jobs interacting with databases etc

Worker nodes must stay on a local network anyway

If there's an open network (for all the nodes), interactive jobs can run
already now - perhaps a new RSL atribute is needed.

 Guest VO
==========

stays the way it is, keep a separate VO for guests. Whether to accept them
or not is the local policy

 Other Grid projects
=====================

Globus : 2.0 is stable, 2.2 is unstable in the MDS part; OGSA alpha - next
year

GRAT - simple scripts, GRAPPA - Java CoG based portal, submits via globusrun

MAGDA - stores manipulates database information; moves files using some 
protocol. Very raw.

EDG: something works, but not quite

 Job log
=========

stdlog : new attribute, analogous to stdout

 Data management
=================

overview of WP2 tools - Oxana and Aleksandr

 Job management
================

issue of jobs restarting by PBS: out of control, better to have GM taking
care of it (no resubmissions by PBS).

important: jobs can not overwrite files, only a tool ran by the Production 
Manager

how do we run pile-up jobs ...

should be possible to have output directory name equaling to the job name

file-based brokering: add a weight \sum{size_i}{flag_i} 

 DC1
=====

3-4 TB of disk space promised by Farid; no other problems apart of the disk
space are foreseen

 Meeting in Uppsala
====================

closed or not? Shall we advertise it widely or not? Ask steering commitee

Technical NorduGrid meeting to be held in parallel with the LHC workshop.

 Data management
=================

File replication
 - 

Logical dataset greplication

Storage Resource Broker - functionality explanation by Balazs

Meeting with WP2 people - Oxana to discuss with WP2 people

Aleksandr: ideas about changing schema for the replica catalogue. Balazs
will write the schema.