Virtual Parallel Languages and Environments

Geoffrey Fox NPAC
Andrew Sherman SCA
Cal Rbbens Virginia Tech
Manish Parashar Texas

Sherman SCA -I

* Agrees with my analysis and goals with different technology
* Virtual Shared Memory very good for PSE
* Store data so greater than one process can examine at same time -- enables computational steering and monitoring simultaneously

Sherman SCA -II

* Khoros-Paradise System for Navy -- Khoros natural as Navy interested in signal processing
* Paradise replaces transport mechanism in Khoros and you have virtual shared memory -- not file -- between modules
* Piranha does scheduling of different modules including replication of the different stateless(often true) Khoros modules
* This enables parallel Khoros

Parashar -- Black Hole

* Process involved complex cycle between CS and application scientists to identify requirements
* Merged multipole, black hole and FEM technologies
* DAGH Technologies for parallel adaptive hierarchical AMR implemented in C++
* Has shadow grid hierarchy for on the fly error estimaion
* Good performance compared to hand coded software
* C++ driver calls Fortran routines in coarse grain model

Ribbens

* Was in ELLPACK project at Purdue -- a user not developer of languages
* Works with engineers at Virginia Tech
* There is no holy grail language or paradigm
* Some sacrifice in raw performance possible
* MDO codes need shell scripts/wrappers

Ribbens

* Compares with Pasadena1 workshop

- likes HPF and MPI forum process
- Not clear if good progress in tools
- What about process of building "advanced prototype"

* Is there a PSE forum -- is problem well defined
* Sems to agree Web is a reasonable starting point
* Agrees that AVS is a good model and in general visual programming environment
* What is a VPE (Virtual Parallel Environment)
* How important/useful is:

- visual programming
- shared address space
- a template
- existing code re-use -- one trusts existing code!

Questions from Audience -- I

* Why shared memory in Sherman's example

- User sees conventional Khoros
- Systems uses shared memory

* Ribbens says shared memory supports incrementalism
* Only developers or expert users will address closely coupled applications
* Skjellum -- Java linked to MPI -- abandon PERL

Questions from Audience -- II

* Do we agree on division between coarse grain and fine grain parallelism
* AVS/Khoros checks if modules should be connected
* Is set of modules the correct model!

- Smith says no -- do top down -- he seems to be very puristic as current modules "artificial" -- I don't agree!
- rather mix top down and bottom up approaches

* PDE's solvers are clearly used cf. NASTRAN but PSE"S are not taking world by storm
* Typical discussion of interdisciplinary research with industry, computer science -- application interaction, computational scien