We combined the answers the above questions because there
was a great deal of duplication in the answers. NERSC
responses to these comments are included in bold.
[ BACK TO TOP ]
- Provide more cycles / more machines: 26 responses
- 10 - more vector cycles; keep the C90
- "Enabling truly large-scale computing. A vector machine with tons of
memory devoted entirely to HUGE jobs!"
- "Massively parallel vector machine. At least 1024 processors."
- 8 - more T3E or MPP cycles
- "Our research depends very much on HPCC technology. More computing
resources available to us will allow us to complete century long
simulations and comparisons."
- 7 - more cycles in general
- 1 - SMP
- "More emphasis on SMP architectures. I fear the MPP approach may be
dead in a few years and is not relevant to many computing projects
in DOE."
NERSC is addressing this concern through the NERSC-3 procurement process.
[ BACK TO TOP ]
- Better software support: 20 responses
- wanted for the T3E:
- multithreaded Perl
- full ports of CERN libraries
- GNU tcsh
- full ssh to/from with authentication
- Concurrent Versions System (CVS)
- more end-user oriented tools like PSPASES, pPCx, optimization tools
- latest version of emacs
- improve ease of use
- wanted for the PVPs:
- 3 requests for keeping cf77
- f77 to f90 translator like VAST
- faster debugger than totalview
- MOLPRO for chemistry
- CERNLIB
- tcsh
- ezget libraries from LLNL
- wanted in general:
- increase variety of supported software
- better support for fixing Cray Fortran bugs
- better file manager and editor
- ghostview
- pop mail server on SAS, email on SAS
- a set of Unix scripts on line for making batch file submission easier
and more user friendly
- in-house software mode as needed to improve use of systems
Some of the software mentioned above is already on the machines, for
example: tcsh, CVS, emacs.
We can't provide
some of the software listed because it is no longer supported, for example, f77. We
evaluate all software requests, and consider the anticipated usage and
the cost.
[ BACK TO TOP ]
- Better / different batch management: 19 responses
- 8 - different T3E queue management (of which 5 for longer queues
and 2 for faster debug turnaround time)
- "I am wondering whether the T3E scheduler could predict when a given
job in the queue would be expected to start execution (this
prediction could perhaps be included in the qstat listing);
this prediction of course would be subject to change, but might be
helpful for planning."
- "you have to figure out some way that it is possible to do medium
sized development work in a acceptable time frame. some bugs show
up only if the job is a little larger - and in the present set-up
with about 2 jobs / day ... it takes for ever to find anything."
- 7 - better batch software
- 4 - different PVP queue management
- "give more priority to large job which can not be done locally"
NERSC has developed a new proposal for the T3E queue structure which should be
implemented by December 1998:
queue name current limit new limit change
----------------------------------------------------------------------
serial 4 hours 4 hours create separate pipe
debug 0.5 hour 0.5 hour create separate pipe
pe32 4 hours 4 hours no change
pe64 4 hours 4 hours no change
pe128 2 hours 4 hours 2X change
pe256 1 hours 4 hours 4X change
pe512 1 hours 4 hours 4X change
gc128 12 hours 12 hours no change
gc256 12 hours 12 hours no change
long_128 - 12 hours new queue
The J90 Cluster queue structure will be evaluated next. We will try to make the NQS
software easier to use and understand by writing better web documentation.
[ BACK TO TOP ]
- Better/different allocations process: 11 responses
- make it easy to get increases during the year
- better notification to users (not just PIs)
- make allocations more easily available to local users
- simplify / rationalize process
- need larger T3E allocation
- change and make more flexible
- ERCAP should be replaced - need to be able to use powerful editor
like LateX
- earlier deadline this year was a major inconvenience
- make allocations based on science, not whether DOE funded
[ BACK TO TOP ]
- Improve documentation/training: 10 responses
- make info easier to find
- improve writing style
- more info
- better info on using math and graphics libraries
- better batch documentation
- need Unix tutorial
- want manual and tutorials in hand
- more high-end training / computational science support
- more remote site training
- better communication of machine status
- better communication of available software
We do continuous work on web pages; the survey tells us we need
to focus especially on the software pages and batch documentation (also
on the visualization pages, but these affect fewer users). We will
provide a new search engine by the end of January, 1999.
We are now offering teleconference classes:
see NERSC Training Classes.
[ BACK TO TOP ]
- More storage, better storage interfaces, better file management, new file services: 9 responses
- 3 - better disk space management (guarantee a few megabytes per
user permanently on-line; migration system unreliable; home
directory policy keeps changing; more space in the work disk)
- 2 - better user interfaces (improve or replace HPSS; put more
effort into system software development; CFS is brain-dead)
- 2 - more shared file systems (access to mcurie from newton;
access to vis server)
- 1 - more capacity
- 1 - tape services ("we have many GBs on DLT tapes")
HSI is being introduced as an HPSS user interface. We will continue to get more
storage capacity all the time.
[ BACK TO TOP ]
- Better accounting/charging/account management procedures: 8 responses
- 3 - provide better accounting program; setcub hard to use or confusing
- 2 - account creation too slow (1 was for HPSS account)
- 1 - setcub scheme not enforced automatically on the T3E
- 1 - resource allocation should reflect user needs: talk to users in
account before setting up the account
- 1 - better account and queuing policies that allow a group to work as
cohesive unit
NERSC is working on a plan to redesign CUB, the database which manages
accounting information. Deployment of the new system is expected for FY'2000.
More account creation procedures will be automated this year, especially for HPSS accounts.
[ BACK TO TOP ]
- Keep users better informed; better interactions with users: 6 responses
- 3 - keep users better informed about system and status changes
- 2 - be more user oriented (rather than computer systems oriented)
- 2 - inform all users of changes during the allocation award process,
not just PIs
- 1 - improve consultant quality
- 1 - keep users informed about your plans for the future (e.g. announce
CFS to HPSS plans now)
One of the things NERSC learned from the survey was that users would like
more email notification from NERSC.
We have already started sending out more
information by email in response to this.
[ BACK TO TOP ]
- Improve the T3E / other T3E issues: 6 responses
- 2 - keep pierre separate
"My guess it will lower down overall
efficiency and realability."
- 1 - resolve T3E performance issues
- 1 - get ssh
- 1 - make it easier to do development work
- 1 - more flexibility in allocation of memory and inodes
[ BACK TO TOP ]
- Faster networks / better response time: 5 responses
- 4 - speedier network access
- 1 - improve ftp speed on Killeen
[ BACK TO TOP ]
- Less down time: 4 responses
- 2 - for T3E/HPSS
- "Twice-weekly mid-day T3E downtimes / HPSS downtimes were a nuisance."
- 2 - less downtime in general
[ BACK TO TOP ]
- Provide a workstation farm; better PDSF support: 4 responses
[ BACK TO TOP ]
- Visualization support/software: 4 responses
- 1 - improve graphics support to the average user
- 1 - improve access to the vis server
- 1 - provide g-sharp or similar software
- 1 - provide readily available 4-D graphics and animation
[ BACK TO TOP ]
- Increase research / scientific computing support: 3 responses
- Need more people in scientific computing.
- "Have experts in areas of user interest available as resources."
- "Add to research computing facilities to compliment the production
facility. Explore innovative use of non-numerical hardware and software,
e.g. visualization environments."
[ BACK TO TOP ]
- No changes needed: 3 responses
[ BACK TO TOP ]
- Miscellaneous: 3 responses
- interoperability with other systems (workstations, etc.)
- X server
- "My principal aggravation is in being asked to provide research
documents to NERSC..."