|
Support for Collaborative Work and Scheduling Assistant for Parallel
Applications
Dan Grigoras
Technical University Iasi, Romania
Hosted by Ian Foster
10:30 AM, March 2, 1999
Building 221, Room A-216
|
| Abstract |
Large heterogeneous distributed systems are currently in use and tend to
play a more and more dominant role. Such systems are characterized by huge aggregate
computing power and storage, diversity, availability and suggest new computing paradigms.
At the application level, collaborative systems can help organizations in becoming more
competitive and in saving corporate knowledge, but at the system level, new problems
arise. Dealing with heterogeneity and making full use of available resources is not a
simple task.
My presentation deals with a software product, under development within our research
group, Internet Works 1.0, and a scheduler assistant for parallel applications execution
on distributed systems. The former piece of software includes full facilities for
collaborative work - whiteboard, video and sound, chat, shared editing, but still has no
facility for knowledge extraction and conservation. The latter started from the idea that
most of the current applications have certain execution behavior, which can be used in
classifying them. Consequently, previous application schedule can be stored and used again
when similar applications are to be scheduled. Two heuristic algorithms are in use, the
Taboo list, and genetic algorithms. Preliminary results are encouraging.
The other research topic of our group is component-based distributed applications design,
which although at the beginning proves some result. |
