Many groups, including NSF, the Alliance User Advisory Committee, and
NCSA's Scientific Computing Division (SCD) staff, all want to know what
users think about the computational resources and services provided by the
Alliance. Each spring a user survey is made available on the web to solicit
comments.
The 2000 user survey, administered in May, was answered by more than 300
users, a jump from previous years when approximately 175 replies were
received. The increase is attributed to an aggressive announcement campaign
waged by Bruce Loftis, who leads the NCSA Scientific Support team and who
was responsible for this year's survey implementation.
"We're delighted so many users took the time to answer our questions. It
really is a terrific opportunity for them to give us input on Alliance
computation environments and on future hardware and software purchases,"
he noted.
Developed and administered by SCD staff, the survey asked more than 40
questions on topics ranging from the hardware, software, and visualization
resources at the user's primary compute site to the quality of consulting
support, web-based information, and allocations processing. Respondents are
also asked to provide information about their research activities, use of
various Alliance resources, and the nature of their computational research.
Evaluation of Computational Environment
Because most of the 2000 questions were identical to those on the 1999
survey instrument, comparisons are possible. The detailed analysis of the data
revealed users continue to be quite pleased with the
computational
environments provided by the Alliance. Productivity barriers identified
include queues and turn-around time, perennial issues raised by users and
constantly reevaluated by staff. Security
procedures are the source of some
complaints, but Alliance and NCSA staff believe that the current requirements
cannot be modified.
Over half of the respondents have run a job using more than 32 processors,
which is consistent with what NCSA staff observed over the last year. A
similar number of respondents use
MPI
for interprocessor communication and
many use several different methods. The use of MPI is consistent with the
strong call for MPI-oriented
training materials in the 1999 survey. (In
September, a new web-based course on MPI developed by the Alliance PACS
partners will be released.) A wide variety of
performance tools and scientific
libraries were listed as frequently used. Scientific support staff will use this
information to develop directed training activities to provide users with skills
to accomplish effective large-scale high-performance computing.
Respondents also indicated other scientific libraries, performance tools, and
commercial application software that
they would like considered for
acquisition. NCSA's new software administrator, Ken Sartain, will be
spearheading the effort to meet as many needs as possible within the
constrained software budget. Sartain will work with NCSA and Alliance
software coordinators in evaluating the requests.
Evaluation of Services and Support
Once again consulting support received high marks with the majority
indicating they are very happy with the support they receive from the
consulting teams. Among the suggestions for improvement was a perennial
favorite: offering a toll-free number for support questions. The cost of
providing such a number will be reevaluated over the next several months.
Users commented on everything from web-based information (as always, the
results were mixed) to their strongly stated desire for multi-year
allocations
(something the Alliance would need to take up with NSF). They showed a
clear preference for training delivered over the web,
which is consistent with
1999 results. Reasons given for preferring web-based training to live training
are travel costs, time, and personal schedule.
In Sum
"Our users are many and varied – in their fields of computational research and
in their levels of expertise. We are constantly challenged to offer to them the
best resources and optimal system conditions. The user survey is a good
'report card' for us, and we use the information throughout the year,"
commented John Towns, NCSA division director for Scientific Computing.
The results of the survey, other than
confidential identifying information, were
made available in their entirety to NSF, the Alliance User Advisory Board,
and the team leaders of Alliance and NCSA groups whose areas of
responsibility were covered by the survey.
