access_UMUWMeeting6Sep2012Notes - ACCESS-NRI/accessdev-Trac-archive GitHub Wiki
===============================================================
Notes from UM Working Group Meeting
-----------------------------------
11:00am-1:00pm Thu 6th Sept 2012
CAWCR BoM Boardroom, 9 East.
DRAFT 02 by azs, 2/10/2012
===============================================================
Attendees:
Yix (phone),T.Pugh, Scott, Say, M.Rezny,
Mjn, M.Dix, Lwenming, Ilia, Joerg,
Azs. Apologies: Rab
Discussion topics (From previous notes):
1) UM components progress
2) TIDG Technical Infrastructure Development Group status
3) NCI issues
4) NECTAR Updates
5) non-ACCESS components such as OPSUI, use of accesscollab
server etc.
The above was not really followed and we tend to seamlessly
move through the issues.
ADMINISTRATIVE
--------------
i) azs volunteer to keep minutes for this meeting. Will seek
volunteer for next or future meetings.
ii) Subset of this group, T.Pugh, Rab, Mike Rezny, Mike Naughton,
and Martin Dix make up the Technical Coordination Subgroup
which shall next meet on 20/9/2012 and alternate fortnightly
with UMWG. next UMWG meeting will be called for early Oct
( ?? 11am Thu 4/10/12 ??).
iii) CSIRO have engaged new staff: Ian Campbell with
Computer Science background and Python programming
experience to work on ROSE. It may be beneficial
to include him in the UMWG.
Unified Model (UM)
==================
1) AMIP run and coupling of vn8.2 is being looked at
with the aim of making a standard job package available.
Among the considerations is that there have been significant
changes to the UKCA package. Martin Dix doing preliminary
investigation.
2) PS30 NWP package is available at BoM. Not yet looked at.
3) RMIFDF - Project to remove #ifdefs from the UM source
is progressing well. The first set of patches unifying the
source for standard and EndGame dynamics have been sent
to UKMO (Glenn Greed) and he have been able to test it
with little trouble. Second set of patches have now be
sent.
-- Glenn will be presenting on the RMIFDF project to the
UM Project Board. It represents major change to the UM
source so may encounter resistance. Merging with other
development on top of vn8.2 may not be trivial.
-- UMWG was not previously aware of the existence of the
UM Project Board. Being co-UM developers, should we
have representation?
-- A write-up on the RMIFDF project have been done and
updated on UKMO Collab Wiki.
-- This project represents one of CAWCR contribution
to the UM collaboration project.
4) Segmentation Fault issue solved by "Bounds Check" compilation
of offending routine.
-- This issue have not been fully resolved since the workaround
using "Bounds Check" compilation work well enough.
-- Tim suggest the issue may be related to a compiler bug.
-- M.Rezny commented that the issue need to be documented
for common benefit, which raised the question that
currently we have not identified site for these type
of information. Mjn sugggest CollabWiki newsgroup.
-- Ilia plan to investigate the bug further at some stage.
5) As per previous meeting notes, UM8.2 standard jobs are
available, but further work is required, and will be ongoing
to further refine and expand what is available.
TIDG
====
Technical Infrastructure Work Package.
UM Partner's Board Meeting have taken place.
CAWCR's input to the work package have been sent back to
UKMO and is now awaiting UKMO's response.
ACCESS, UM SUPPORT and related
==============================
1) um_help and access_help email addresses are aliases of each
other and managed using NCI's RT system. "cable_help" currently
also go to the same list. "nci help" and "cms/climate_help"
are other help email addresses.
M.Rezny reminded the group that we are using the RT system
as guest of NCI. Rita, Margaret and Judy have provided valuable
help but we need to be good "citizens". We should close tickets
ourselves which require using the RT web interface.
Careful usage of email "Subject lines" and removal of
unnecessary "email tails" is desirable. All correspondence
are archived by the RT system, "email tails" are generally
redundant information. Always ensure the correct string
in the form [nf.nci.org.au #NNNNN] required for proper
handling by the RT system is in the subject line.
[-> Separate different issues into different subjects]
M.Rezny will organise training in using the RT system
some time after 3rd week Oct, when Robin Bowen is around.
M.Dix/T.Pugh: Robin has been tasked to develop BoM's
help/support system for the NCI environment. Quite likely
it will make use of the available RT system.
NCI Issues
==========
1) NCI have asked us to "slow down" and make proper plans before
proceeding .
Ben Evans believe that our requirement for "accessdev" serve
can be met using the "cloud".
T.Pugh stated that (as per NCI request) we need proper
documentation of our requirements and plans.
Requirements include:
i) Server Hardware -- to be met by virtual server in the cloud.
ii) System Administrator -- NCI and/or BoM/CSIRO/CoE staff
iii) Operating System / Software stack
iv) MARS, Mass Store / Archiving requirement.
Xiao have written up her requirements and sent to relevant
people.
Request for particular s/w packages, versions etc will
be addressed later.
2) A meeting have taken place addressing the requirement for MARS
at NCI. Likely solution is a disk-only MARS utilising openDAP.
Tan Le is looking at the implementation with time-frame for
availability targeted for January 2013.
Of note: At BoM, 3-4 Tb of data is archived to MARS everyday
so current capacity of BoM MARS is probably in the Petabytes.
3) The disk /project/access disk on vayu hit the quota limit,
though quota appeared to have been increased from 2GB to 40GB
at some stage. There are large amount of CMIP and other
data which probably can be moved elsewhere. CSIRO staff
working on CMIP have just returned from holidays and will
be consulted. [An increase in quota is desirable nevertheless
for migration work to proceed ]
There is need to create /project/access/sms and
/project/access/agreps. We may also want to have
/project/access/bom for BoM's legacy systems, but
in general we want common structure for the access
community.
Should we move /project/access/ to /home/?
Little interest in pursuing creation of "/access/" link
to "~access/". Currently "~access/" is set to "/project/access/"
People who have management permission to "~access/"
are Martin Dix, Mike Rezny, Margaret Kahn, Scott Wales
and Say.
For solar, BoM have a "Change Management" process in place
but we probably do not want/need similar arrangement.
We also do not want a few months gap between request
and action.
Quotas for individuals and project will require further
discussion.
4) Currently BoM access users tend to have ACCESSDIR
environment variable defined in their ".profile".
It was suggested that at NCI, we should use "access module"
instead. Module environment have
been set up/ being developed for ROSE, cylc and SMS.
ROSIE is not available and it is not appropriate to
install it on vayu, nor accesscollab. The new
accessdev server is most likely the best place for it.
It was suggested that we need to request/get this
server ASAP. Martin Dix, Mike Rezny will address
this together with Ben Evans (BoM staff ?? ).
Any request for S/W packages on the server will
be address later
5) Shared Repositories
Access Control List (ACL) management of the repositories
on access-svn is through trac interface.
A few people have been designated as "admins". It is believed
that small group of admins is desirable although backups
would be required when people go on leave.
To check if azs is a member of the UM repository admin group
with permission to update its ACL.
6) UM Tuning.
Enda's contract is now complete have he have submitted his
report.
Tim Pugh will forward Enda's report to Group members. Also
available and will be forwarded are:
- Paul Selwood's response to Enda's report
- UKMO IBM P7 report, which Paul Selwood decide to share
with us in response availability of Enda's report.
The report address items such as "rework of IBM's vma's"
"removal of IBM devs", "blocking improvements", OpenMP,
I/O servers. OpenMP is still an issue, as is hyper-threading.
- IBM P7 supercomputer is expected to go operational at
UKMO after the paralympics.
Further collaboration with Intel in the future is under
consideration. Collaboration with Fujitsu is taking place
with a meeting scheduled.
It is desirable perhaps to combine the collaboration
with Fujitsu to include Intel, since we will still
be using intel'l CPUs and compilers.
Other potential partners are CoE and CSIRO eResearch.
Workshops to be sponsored by Intel and Fujitsu
Meeting ends: 1:00pm
Next Meeting: 11am-12:30pm Thu 4/10/12.
=========================================================###### =========================================================
[azs, 2/10/2012] Transfer notes from email 6/9/2012