Notes from the Conference Call Clustervision/IBM/University of Birmingham
9 September 2009
Present (on the call)
Paul Hatton, University of Birmingham [PSH]
Gerd Busker, Clustervision [GB]
Alex Ninaber, Clustervision [AN]
Antoine Schonewille, Clustervision [AS]
Jim Roche, IBM [JR]
Mark Dunham, IBM [MD]
Apologies
John Owen, University of Birmingham [JGO]
Also possibly mentioned in the notes
Peter Watkins, University of Birmingham [PW]
Alan Reed, University of Birmingham [AR]
Aslam Ghumra, University of Birmingham [AG]
Notes from the meeting
-
The notes from the previous meeting were accepted without any corrections.
-
The recent filestore and other, maybe unrelated, issues were discussed,
mainly from the point of view of communication between all 3 parties
involved (Birmingham, Clustervision and IBM) rather than in-depth technical
issues. The main points to emerge were:
- a round-table meeting at Birmingham to discuss information flow would be
valuable all round. AN is planning to visit the UK in the next few weeks;
this should be scheduled whilst he is here
Action AN: circulate proposed date(s) for a debrief at Birmingham
- AN clarified that Clustervision and IBM did not give conflicting advice
at any stage; PSH accepted this
- PSH was unclear on who was responsible for ensuring that all parties
were kept informed of advice and suggestions received from Clustervision and
IBM. AN suggested that Birmingham should have ensured that this was the
case, including forwarding all mails received from, and sent to, IBM to
Clustervision but this needs to be clarified in the debrief. Once PSH
realised that not everyone was seeing all of the mails he asked AR and AG to
ensure that this happened.
- a timeline of events as seen by Clustervision and Birmingham would be
valuable to inform the debrief
Action AN and PSH: prepare and circulate independent timelines to
inform the debrief
- AS said that there were several distinct and maybe unrelated issues that
arose at the same time.
- AN said that some of the issues would have been noticed by AR and AG if
the cluster had been monitored more closely. PSH suggested that a one-day
on-site training from Clustervision to AN, AG and PSH on monitoring the
cluster, including the time commitment that is needed, would be valuable,
and asked for a quotation for this.
Action Clustervision (not sure who): prepare quotation for one-day
on-site training for cluster monitoring
- JR has discussed our requirements for shared project filestore with Andy
Parker at IBM, who wouldn't recommend the use of GPFS ACLs. The schedule for
the regular filestore checks and GPFS update meetings/discussions is yet to
be arranged; JR is in conversation with Chris Brown at IBM about this.
Clustervision would be welcome to join in these meetings.
Action JR: set up
and circulate schedule for 6-monthly filestore checks and meetings. The
first one is tentatively proposed for the third week of January 2010.
Action JR:
circulate Andy Parker's suggestions on managing shared project areas
- The DCV server has been delivered. The graphics card has been delivered
but the riser card(s) to supply the power from one of the PCI slots are
still awaited - there are no spare connectors to the power supplies in the
server. XP is the preferred O/S, due to it handling OpenGL better than
VISTA/Windows 7; JR pointed out that VISTA includes downgrade rights to XP.
PSH will follow this up.
Action JR: source graphics power riser
card(s)
for the DCV server
- The next User Forum will be on 24 November 2009. The afternoon will be
given over to the Meet-the-researchers event, whereby selected research
groups, following a general announcement and submission of a half-page or so
outline of their work, would be invited to a meeting lasting around 45
minutes to an hour with IBM to discuss their work and possible involvement with IBM
- such work would need to be capable of running on BlueBEAR, possibly with a
path up to larger resources. IBM delegates would include MD, JR and a
Hursley technical representative; Clustervision would also be welcome.
Sponsorship for the lunch would be welcome.
Action PSH: mail out invitation to
the User Forum, highlighting the afternoon event
Action JR: invite a Hursley
technical representative to the event
Action GB: discuss sponsorship of the event with Matthijs van Leeuwen - on
past costs, around £7 per head would be good
- Rolls-Royce are keen to visit VISTA - they have other priorities at the
moment, though.
Action JR: confirm date for RR VISTA visit
- PSH suggested that some relevant demonstration material, especially if it
could be displayed in a stereo-capable application such as Avizo, would be
useful both for the potential RR
visit and in a broader context, maybe to showcase work that is going on
elsewhere using IBM systems.
Action JR: seek out demonstration(s) for VISTA
- The cluster usage figures would be of interest; PSH will circulate
these.
Action PSH: circulate usage figures
- Cluster-specific MPI training has been previously discussed. PSH has built
some applications such as ScaLAPACK and has succeeded with an OpenFOAM
parallel build following advice from Aleksander Korzynski before he left Clustervision, along
with a great deal of trial-and-error. The need for such training is probably
diminishing, although PSH feels that such training would have been very
beneficial when the cluster was installed.
Action GB: discuss cluster-specific MPI training and/or documentation
with AN
Next call
9.30, Wednesday 23rd September 2009