Title: Department of Computer Science Seminar
Date: Tuesday, Dec. 12, 2000
Time: 9:20 am to 10:20 am
Venue: Room N101, CSIT Building [108]
Speaker: A/P Harold CASTRO (Ing. Sistemas y Computacion, UNIVERSIDAD DE LOS ANDES )
Description: "High level parallel programming on heterogeneous computer networks"


                        Abstract


Working on parallelism is becoming a difficult task for many
 universities, particularly those in the third world countries.  
Hardware and softwares high cost is the main obstacle.  However,
universities are the most interested institutions on working in
 parallelism, not only to execute some heavy applications, 
but also to offer entire environments to teach and to do research on the subject.

During last decade, several projects have been developed to tackle
this problem.  These alternatives exploit parallelism on computer
networks, reducing costs by taking advantage of
the hardware and software already available in an organization. 
PVM, MPI and P4 are the main examples of this kind of programming
facilities, which define a parallel virtual machine composed 
by a set of computers in a local network. 
Here, each entire computer is seen as a processor 
of the (virtual) parallel machine.

In these environments, programmers main possibility 
is to launch remote processes (with some kind of synchronization) 
and to establish communication between processes in a parallel
application.  These features are in fact needed 
to express parallelism within applications, but they answer 
to very restrictive models of parallelism and burden programmers with operation
details, which should not be at their charge.  

What we present in this talk is a parallel platform, Mapaná, 
currently being developed at Universidad de los Andes. 
Mapaná offers a complete development and execution environment 
of parallel programs, based on a high level programming model, which clearly
separates programming and implementation on parallel environments.  
Besides the traditional services, Mapaná has several 
features not included in other virtual machines: global memory,
only one executable file for parallel application, 
parallel file system and binary code compatibility 
through different systems.


 
URL: http://cs.anu.edu.au/lib/seminars/seminars00/dept20001212