http:^^www.cs.cornell.edu^info^people^nikosc^cs_projects^index.html

来自「This data set contains WWW-pages collect」· HTML 代码 · 共 135 行

MIME-Version: 1.0
Server: CERN/3.0
Date: Sunday, 24-Nov-96 22:57:13 GMT
Content-Type: text/html
Content-Length: 6359
Last-Modified: Thursday, 21-Dec-95 19:08:30 GMT

<TITLE> Masters of Engineering Projects </TITLE>
This year all our projects will address issues associated with 
the development of an environment for designing, specifying,
implementing  and testing Adaptive Mesh Refinement (AMR)
methods for tackling very complex problems (ex.  3D spiraling 
coalescence of two black holes) from computational sciences. 

The participants in the development of this environment will have 
the opportunity to work on very specific projects that can be developed
individually, but at the same time will require some degree of interaction 
in the design, validation and testing phases. Thus, in addition to the 
opportunity of acquiring specific skills and knowledge by working 
on a very narrow and interesting problem the opportunity to see and 
understand the procedure of designing and implementing a complex 
software for parallel and distributed platforms will be given.

Projects :
<ul>
<li>  <A href="#project1"> Intelligent Graphical User Interface for AMR </A> 
<li>  <A href="#project2"> Parallel Runtime Support System for AMR </A> 
<li>  <A href="#project3"> Dynamic Load Balancing Algorithms for AMR </A> 
<li>  <A href="#project4"> Parallel PDE solvers using AMR methods   </A>  
<li>  <A href="#project5"> Parallel structured grid generation for 2D/3D complex geometries  </A>   
<li>  <A href="#project6"> Scientific computing benchmarks for multithreading on parallel and  distributed platforms </A>  
<li>  <A href="#project7"> Study the impact of restructuring CG scalar algorithms and codes in terms of CG's performance on MPPs </A>  
</ul>


<A name="project1">
<H3> Intelligent Graphical User Interface for AMR (GUI-AMR) </H3> 
This project requires the design and implement an intelligent
   frond-end for specifying 2D geometries, operators and various 
   parameters required to solve numerically, time-dependent PDE problems. 
   This project consist of two components:
	A. Problem specification 
        B. Visualization
   Existing Problem Solving Environment (PSE) PELLPACK will be extended.
   This project is recommended for students with interest on GUIs,
   PSEs, I/O, and in general front-ends for high-performance computing 
   environments.
</A>
<p>
<A name="project2">
<H3>Parallel Runtime Support System for AMR (PRTS-AMR)  </H3>
This project requires the design and implementation
    of communication and threaded modules requirted for
    the efficient implementation of data-movement and 
    control primitives for parallel AMR methods. 
    Existing communication software like MPI, AMs and threaded
    packages like Qt and PORTS0 will be utilized.
    This project is recommended for students with interests
    both on systems, parallel I/O, parallel compilers and computational 
    sciences. An interaction with Bernoulli and Split-C projects
    is expected.</A>
<p>
<A name="project3">
<H3>Dynamic Load Balancing Algorithms for AMR (DLB-AMR) </H3>
This project requires the development, implementation and evaluation of 
   of dynamic load balancing algorithms for AMR: (i) direct like generalized 
   spectral bisection and its derivatives and space-filling curves 
   and (ii) incremental. The objective is to develop (i) a DLB-AMR module
   for PRTS-AMR and (ii) acquire useful knowledge in solving an instance 
   of a very difficult combinatorial optimization problem: dynamic load
   balancing of adaptive computations on a network of time/memory-sharing
   heterogeneous workstations and MPPs. Existing algorithms and software
   will be extended and new ones will be build. This project is recommended
   for students with interests on the solution of practical optimization 
   problems related to parallel and distributed computing. </A>
<p>
<A name="project4">
<H3>Parallel PDE solvers for time-dependent problems using AMR methods</H3>
This project requires the implementation of a library of routines for the 
   discretization and solution of 2D/3D wave equation on MPPs and SMPs. 
   Both non-threaded and multithreaded paradigms will be used and evaluated.
   In addition accuracy, and stability of multithreaded computations will be
   analyzed. High-performance languages like HPF and low level like 
   Fortran/C plus message passing will be used. This project is recommented
   for students with interests in parallel numerical computing, and 
   scalability analysis. An interaction with PRTS-AMR project is expected.</A>
<p>
<A name="project5">
<H3>Parallel structured grid generation for 2D/3D complex geometries</H3>
This project requires the development of parallel algorithms and 
implementation of a library of parallel grid generation modules using :
<ul>
<li> Algebraic Methods
<li> Elliptic Methods
<li> Moving Algebraic Methods
</ul>
High-performance languages like HPF and low level like Fortan/C plus 
message passing will be used. This project is recommented for students
with interests in parallel numerical computing. Existing state-of-the-art 
scalar algorithms and code will be used. An interaction with WLIB-AMR is
expected. </A>
<p>
<A name="project6">
<H3>Scientific computing benchmarks for multithreading on parallel and 
   distributed platforms</H3>
</A>
<p>
<A name="project7">
<H3>Study the impact of restructuring CG scalar algorithms and codes in terms of CG's performance on MPPs</H3>
This project requires the development of a Conjugate Gradient algorithm
suitable for multiprocessors and includes the implementation of CG using
Typhoon parallel compiler. Issues related to expensive dot product 
operations will be studied:
<ul>
<li>  How much the algorithm restructuring can help in improving
      performance compared to well known CG algorithm?
      Is it worth the effort and possible increase in space complexity?
<li> How much the code restructuring can help in improving
     performance compared to straight forward cg code? 
     Is it worth in doing the extra step in exploring functional
     parallelism on top of the data-parallelism?
<li>  How much will cost to use multithreading in (i) 
      implementing functional parallelism (ii) masking some of the 
      global address space overheads? Is the overhead worth the 
      increase in software complexity? What about error and stability?
</ul>
</A>
<p>

For more information contact:  <A HREF="http://www.cs.cornell/Info/People/nikosc.html">  </A>