Minutes of NAF User Committee meeting from 13.1.2010
----------------------------------------------------

Present: Steve Aplin (ILC), Wolfgang Ehrenfeld (ATLAS), Kai Leffhalm
(NAF), Angela Lucaci-Timoce (ILC), Andreas Nowack (CMS), Harmut Stadie
(CMS)

Excused: Jan Erik Sundermann (ATLAS), Andreas Haupt (IT), Yves Kemp
(IT)

1. News from the chair:

The NAF web documentation was partly rewritten and
restructured. Everybody is welcome to have a look and give feedback.


2. Status report:

Status report was given by Kai. See the agenda for the report. Below a
few highlights from the discussion are listed:

 - Lustre quota

Quota per user can be set per Lustre mount by the VO admins. This
works for Zeuthen Lustre instance and after the January shutdown also
for the Hamburg Lustre instance. By default there is no quota
set. Even if no quota is set, the tools will report the current user
usage, which was appreciated by all experiment.

From the NAF side there will be no sub groups, which are needed for
group quota. Some of the experiments envision the use of group quota. 

For example CMS would like to use half of the Lustre space for users
and half as data cache. It is not clear how this can be achieved with
user quotas alone and a growing number of users. ILC explicitly asked
for group space for a few working groups. The anticipated solution
from the NAF side with user quota is the following. One user becomes a
power user with larger quota and is then responsible to fill and
administrate one group space. Is might work for some time but is not
really scalable. ATLAS always asked for group quota and not user quota
as ATLAS wants to split the Lustre space into two partitions with
different file life time. This is not feasible with user quotas and
ATLAS is not happy at all with the current development.

One solution is to do some partitioning of the Lustre space by
splitting up the Lustre instances. This would avoid creating sub
groups but it a rather rigid solution. 

 - Lustre cleaning

The NAF will not provide any tool for automatic clean up of files in
the Lustre space.

ATLAS ask for this many times and request sudo rights to do the
cleaning on their own.

 - AFS scratch space

It is planned to give every user 10 GB of AFS scratch space (no
backup), mounted into the home directory as ~/scratch. As soon as the
NAF people find time they will implement it for new accounts and
migrate the old ones.

 - glite

For SL5 glite 3.2 is installed and can be accessed via ini
glite32. glite 3.2 is mainly a 64 bit build for SL5.

ATLAS and CMS need glite 3.1 for they Grid tools for some time. ILC
will do some tests with the new glite version.

 - group profiles

ATLAS asked explicitly for group profiles as there are a few use cases
which are not very practical with the ini setup. The NAF will check on
the status of the implementation.


 - ATLAS CMT update

The NAF provided for ATLAS an AFS patch from CERN, which should give
better performance for cmt. This was seen by ATLAS but to the nature
of the patch the NAF didn't want to install it on more than one
machine. Forcing all cmt user onto one machine will give usually
worse performance than splitting them up over three machines with
some worse performance. Therefore this patch will not be used.

Another option is software installation into the local disk, but
clearly not into /tmp. Will be followed up offline.


3. SL5 migration of work group servers

During the January shut down more of the work group server should be
migrated to SL5 and it was agreed that the login hosts should point to
SL5. All experiment requested 2 SL5 work group server and one SL4 work
group server. For ATLAS this should be 2 SL4 work group server.


4. Action items:

Many action items were touched but not finally closed. The list will
be updated at the next meeting.


5. AOB:

k5log should be used instead of klog for creating an AFS token for
external AFS cells. There are no restrictions as before with klog.

ATLAS and CMS gave very positive feedback to the NAF during the time
of first LHC data in 2009. CMS was able to run already after a few
hours after data taking over the data on Te NAF batch system with out
any problems. The turn around time for ATLAS was not that good but
data was analysed successfully at the NAF.