Now that the new configuration has been exploited by a variety of users, we would like to 
begin developing policies and procedures to enhance its value to the user community.  In 
order to do this, we need to garner some understanding of how the system is being used 
and how it is performing for your work.  We think that a variety of new options are 
possible, from consulting services to optimize programming to training sessions to 
courses and workshops.  We also can change policy used in running the machine. But to 
find the best of these options, we need to make thoughtful assessments of performance to 
date and problems that have come to light.  For us to do this, we must ask you for some 
thoughtful details on your experiences.

We would like you to:

a) Update (if it has changed significantly) description of Science and Computer 
Request. This was originally done using form:
http://www.csit.fsu.edu/facilities/supercomputer/account_application_form.htm 
We have made a (private) web page of current documents at 
http://www.csit.fsu.edu/facilities/supercomputer/applications/     
b) Please add a computational science discussion. This could include discussion of:  
1) Nature of code -- e.g. "Three dimensional Finite Element Adaptive Mesh"
2) "Nature and Quality" of algorithm technology used e.g. "Currently using 
Gauss Seidel, working on switch to Multigrid" or "Using IBM Library for 
Matrix Solver"
3) "Nature and Quality" of Software Technology e.g." good MPI   
implementation". Discuss both intranode (OpenMP between 4 CPU's) and 
internode (MPI) parallelization.
4) Parallel Performance Evaluation: please give us the "efficiency" as a 
function of number of nodes for relevant sizes of problem. This is 
important and please again discuss both intranode (4 CPU's on a single 
node) and internode (across multiple nodes) performance. If a range of 
problem sizes important in production runs, discuss the relevance of 
performance to this and indicate those sizes, which will use most of SP 
time, e.g. "We choose to run so that parallel efficiency is 80%. This 
implies we will use 32 CPU’s for half and 8 CPU’s for other half of jobs".
5) Single CPU performance. Any comparison of SP CPU with other 
machines
6) Any I/O, Visualization, Systems Software or other non CPU issues
7) Any additional comments on policy, management, use of SP by others etc.
Here are some questions you might answer regarding system policy
? How many jobs should users run in each queue simultaneously?  (1 
job), (2 jobs), (any number of jobs)?
? How many jobs should a user be allowed in the queue simultaneously? 
(1, 2, x, any)?
? Should details of users jobs be accessible to the community? For 
example, should "llq -l" be available to all users?
? Should usage statistics be posted on the web for all to see?
? Finally tell us your experience as to abnormal running or waiting times 
on SP.

c) Kyle Gallivan is teaching Computational Science again this semester. Please 
encourage your students to attend. Comment on value, topics (curriculum) of 
courses and/or specialized training would be appreciated.

Thank you