Draft HPCC Software Survey

This software survey is designed to quantify need and opportunity for more and/or better software on HPC or Supercomputing Systems. If there is perceived to be a need, the survey attempts to determine both the nature of the need and the recommend approach to meeting need.

Context

A1: Define your area of work

I work in Academia Large Industry Small Business Government

Check All That Apply
A2: Define where you work

I work in US Europe Asia Other

Check Most Applicable
A3: Define your job category:

Job
Title Decision Maker Technical
Manager Researcher Technical
Staff Other

Check Most Applicable
A4: Define the type of computer use that is your main concern

Type of Computing Running Application Developing Applications Developing Tools Developing Systems Other

Check Most Applicable
A5: Please select of one of the following 4 options to describe your job mix. Capacity computing involves large large jobs for which it is impractical to run on modest size machines. Throughput computing involves involves many small jobs which large aggregated demand but where each job does not require a "supercomputer" scale system. Estimate the category below separately for your own use and that of your organizations.

Nature of Computing Needs Mainly Capacity Mainly Throughput Roughly Equal Capacity and Throughput Not Applicable

Your Use

Your Department's Use

Your Organization's Use

I work in	Academia	Large Industry	Small Business	Government
Check All That Apply

I work in	US	Europe	Asia	Other
Check Most Applicable

A6: Categorize platforms where you are working now and where you think you will working 2 and 4 years out. If you have no opinion as to a particular date, then leave that column all blank. Otherwise numbers in a given column will be treated as relative amounts and renormalized to total 100%. A Computational Grid is a geographically distributed set of diverse computing systems integrated by systems like Globus or Legion to solve individual problems. A "Idle PC/WS Flock" is a set of PC's and workstations in an organization, whose idle time is made available to "production through-put" computing via software like Condor. PC Clusters refer to systems of PC's or workstation thoughtfully put together with relatively high-speed networks to act like a parallel machine. A dedicated cluster is aimed at serving needs of a small group; a shared cluster is the same commodity hardware design but many different users as in traditional center operation. One can use a purchase price of $1M to divide the class called Large Parallel Computers from the smaller systems of "department size".

Platforms Used by Time and Architecture	Now	2 Years Out	4 Years Out
Vector Supercomputer
Large Parallel Computer (MPP)
"Department Size" Parallel Computer
Dedicated PC/WS Cluster
Large (shared) PC/WS Cluster
Computational Grids
Idle PC/WS Flock

HPCC Software Situation and Futures

Now we will try to find out your view on the current situation with HPCC Software and explore various futures.

Application Software

Here we categorize your view of application software.

B1: In questions B1, B2, B3 and B4, we quantify the impact of any limitations you have in your ability to run HPCC Application codes: If you feel limited by inability to run application codes, please tell us if hardware or software is mainly responsible. Check the last button if you are not limited by application codes.

My Computational Productivity is impacted by:	Dominantly Hardware Issues	Dominantly by availability and capabilities of HPCC Codes	Impacted Significantly by both Hardware and Software	Irrelevant Question. I am in great shape.

B2: If you feel limited by application software, please quantify this. (Skip question if you have no such software limitations). Question A6 decsribes the machine classes in second part of question. In this question, check all boxes that apply. The lack or insufficient performance of application software is

Affecting use of HPCC systems by	me	My Department	My Organization
Particularly Affecting the use of the following machines	Vector Supercomputers	Large Parallel Computer (MPP)	"Department Size" Parallel Computer	Dedicated PC/WS Cluster	Large (shared) PC/WS Cluster	Computational Grid	Idle PC/WS Flock

B3: Please List Code Areas (e.g. Finite Difference electromagnetic signature or Particle in the Cell astrophysics) where you feel that HPCC application codes need to be developed as a priority
B4: Please list explicit application codes by name where you feel there should be priority attention

Role of Training

B5: Please comment on relevance of Training below

The amount of Training in development of HPCC Codes should be:	Decreased Significantly	Decreased Somewhat	Left As is	Increased Somewhat	Increased Significantly
The amount of Training in use of HPCC Codes should be:	Decreased Significantly	Decreased Somewhat	Left As is	Increased Somewhat	Increased Significantly

B6: Please enter any further comments on training here

HPCC Systems and Tools

C1: NowCategorize your view of current Tools and Systems software situation. this set of questions addresses the quality of software used to produce application software, not application software itself. Please answer each question in following table

Development of HPCC Tools and Systems Software is sink of resources that could be better used elsewhere
Well Better HPCC Tools and Systems Software would be nice but it is technically impossible and/or prohibitively expensive
There may be opportunity to improve HPCC Tools and Systems Software but this is in the noise. The real opportunity is to take existing tools and develop more application code that can take advantage of HPCC Systems
Improved HPCC Tools and Systems Software will make a real difference and it is worthwhile to attempt this
The current HPCC Tools and Systems Software market is too small to motivate any significant increase in commercial interest in it
There is so much free HPCC Tools and Systems Software that it stifles much potential commercial enterprise
The HPCC field is so fascinated by hardware that it will never spend significant effort on developing better software
You wait! Linux and the Linux Software model will solve most of the problems
It may have nothing much to do with HPCC, but greater use of Web Technology will make my use of HPCC machines more productive
It may have nothing much to do with HPCC, but integration of CORBA into software systems I use, will make my use of HPCC machines more productive
CORBA is not high performance and so is not relevant
You blew it. I put below the most interesting oneliner in this area. Please Phrase the question so that your answer to question is yes.	Check Here if your fill in question to left

Next Steps in Software Development

Here we discuss all types of HPCC Software -- Applications, Basic Systems and Tools

C2: Describe the value of attainable (with effort to be described in later questions) improved software could have to your productivity. Here Software means the whole ball of wax -- Applications, Tools, System. Here value is some nebulous increase in productivity. Assume that improved software is to be delivered over next 3-5 years with existing or new programs which we will be try to delineate in following questions.Please click one button in each row.:

Value of Better Software < 10% 10-25% 25% or more No Idea or Don't Care

Importance to me

Importance to my department

Importance to my organization
C3: Will the required improvement in Software will happen anyway without significant changes in either strategy or funding level? If you believe this, please check here and ignore following questions C4-C8.
C4: Check here if you think that no increase in funding is necessary (from any source) but that a change in strategy would have major impact of the availability of HPCC Software. Explain this text area, how you think this can be acheved.
C5: In the past, it had been assumed that some sort of commercial activity would produce the needed HPCC Software. We will discuss other aspects of this later but here we ask if this could be funded by increases in user purchase price of HPCC Software. In ordinary to quantify this, we ask how much extra you would be willing to pay for software as a function of hardware cost.

For every hardware HPCC dollar, it is plausible to surcharge the listed fraction to pay for additional HPCC Software None 1-5% 5-15% > 15%

For every hardware HPCC dollar, it is plausible to surcharge the listed fraction to pay for additional HPCC Software	None	1-5%	5-15%	> 15%

C6: Indicate a possible increased level of Government Involvement in HPCC Software Development and Research. Note we separate out development (of supported production quality software} from research. This entry does not assume that this is only mechanism to fund HPCC Software and so this is the Government component of any new strategy. Make the increase over current activities and do not include new funding initiatives (such as those following from PITAC report) in the base.

Is increased Government funding essential to improve HPCC Software?
If you checked box in row above, please indicate annual INCREASE in Government funding appropiate for HPCC Software Development	Millions of dollars per year
If you checked box in first row, please indicate annual INCREASE in Government funding appropriate for HPCC Software Research	Millions of dollars per year

C7: Here we determine your priorities for the use of increased HPCC Software Development Investment. Research is not covered here. The machine categories are defined in question A6. All 21 entries are treated as relative weights.

Relative Weight of new HPCC Software Development Activitites	HPCC Applications	HPCC Tools	HPCC Systems Software
Vector Supercomputer
Large Parallel Computer (MPP)
"Department Size" Parallel Computer
Dedicated PC/WS Cluster
Large (shared) PC/WS Cluster
Computational Grids
Idle PC/WS Flock

C8: There are many possible models for HPCC software development and in the following question, we try to get your opinion as to the best model(s) for any new software activities. We have divided the software into the usual three areas -- Applications, Tools and basic Systems. We have divided commercial activities into system vendors, the user industries and the typically smaller independent software vendors. Direct government development is one row -- either through contracts or via development in their laboratories. There are many Government-Academic partnerships as illustrated by the NSF PACI (NCSA at Illinois and NPACI at San Diego) and DoD HPC Modernization PET program. The final row is called Government-Industry partnerships which is not directly being broadly exploited in HPCC Software at the moment. Here we assume that a commercial base is essential for sustainable software but currently this is not easy in HPCC as the market size is too small to support HPCC Specific products. We assume that Government funding is additive to a basic pure commercial activity to essentially fund the HPCC specialization. As these are all quite vague, we follow with a text area to enable you to give your opinions here. As usual the 18 entries in table are treated as relative weights.

Relative Weight of Business Models for new HPCC Software Development Activitites	HPCC Applications	HPCC Tools	HPCC Systems Software
HPCC Vendor Businesses
Industy using HPCC Systems
Independent Software Vendors
Government through contracts and laboratories
Government - Academic Partnerships
Government - Industry Partnerships

Please add any clarifying comments on your answer to C8 here.

Summary

D1: Now we list some areas of HPCC Software and Tools that we would ask you to prioritize

HPCC Systems Software or Tools	Relative Weight
Cluster Management Systems
Debuggers
Load Balancing Software
Mesh Generation Tools
Visualization Tools
Performance Measurement and Analysis
Parallel Compilers like HPF
OpenMP infrastructure and tools
MPI infrastructure and tools
Node Compilers (Cache Management)
Problem Solving Environment and Web Computing Environments
Computational Grid Support Software
Software Frameworks for Multidisciplinary like applications
Parallel I/O
Linear Algebra Libraries
Partial Differential Equation Solvers
Optimization Software
Your Choice 1
Your Choice 2
Your Choice 3

D2: Which user languages should be supported and with what weight? Each column is separately tiotalled for relative weights

Relative Weights for New Application Code Development Language Now 3 Years Out

Fortran

C++

Java

Other
D3: Any Overall Comments should be placed here

Finally send all these nswers to us!

Thank you

Nature of Computing Needs	Mainly Capacity	Mainly Throughput	Roughly Equal Capacity and Throughput	Not Applicable
Your Use
Your Department's Use
Your Organization's Use

Value of Better Software	< 10%	10-25%	25% or more	No Idea or Don't Care
Importance to me
Importance to my department
Importance to my organization

Relative Weights for New Application Code Development Language	Now	3 Years Out
Fortran
C++
Java
Other