|
Short task description
The main goal of the project is to
construct a system that assures remote access to mathematical software
libraries installed on computers working in Polish scientific
computational centers.
Many scientists use specialized
mathematical libraries like e.g. BLAS, IMSL, LAPACK, SCALAPACK or
NAG in software they create. These libraries are optimized for specific
computational machines. They assure high efficiency of the computations
and allow preservation of software simplicity. Many functions included
in the libraries perform mathematical operations that are basic for
scientific computations, e.g. matrix operations, solving systems of
equations etc. Currently there is a need to install mathematical
software libraries on all machines where the scientists might wish to
run software that uses them. Moreover, using a function from the given
library is now possible within the confines of the given system with
no possibility of automatic computations distribution on multiple
machines.
Benefits resulting from the construction
and putting the system into practice may be divided into two groups:
benefits for its users and benefits for the administrators and managers
of the computational systems.
The users will mainly gain:
- the possibility of using high-efficient, specialized and distributed
computational resources;
- access to functions of mathematical software libraries installed on
any computer included in the providing of access system;
- access will be possible independently of having or not having a user
account on a given computer system - users authorization will be performed
on the providing access system level, not on the given machine level;
- the possibility of concurrent execution of whole tasks or their
specified parts;
- the simplicity of the access to the system and insusceptibility to
failures.
All these factors should increase computations efficiency and to improve
the quality of work with mathematical software libraries.
Computational system administrators will
gain:
- the possibility of economical software license management -
installation of all mathematical software libraries used by users working
in the given location will not be necessary;
- no necessity of setting user accounts on computational systems for
all persons interested in using mathematical software installed on these
systems;
- better hardware utilization (by appropriate resource allocation for
tasks and system load balancing).
R&D works connected with the construction
of the system providing remote access to mathematical software libraries
will cover the following areas:
- NetSolve
and Ninf systems
analysis in order to choose a base solution for the system providing remote
access to mathematical software libraries; the analysis will be composed of:
- theoretical analysis of the mechanisms used in NetSolve and Ninf
systems
- practical verification of efficiency of the specific mechanisms:
installation, adjusting and testing of the systems and analysis of the
results
- preparing a detailed project of the system providing remote access to
mathematical software libraries;
- specifying how to exploit the solutions chosen as described above;
- constructing mechanisms that enable the existing applications to
automatically adjust to the use of the remote access system;
- preparing techniques of including mathematical computational
resources working under control of resource managers into the remote
access system;
- constructing mechanisms ensuring applications using the remote access
system to be resistant to its temporary inaccessibility and mechanisms of
assessing remote execution cost-effectiveness;
- constructing mechanisms extending job scheduling and resource allocation
planning subsystems;
- designing a mechanism of historical data analysis on the basis of:
- research on performance processing models for machines with different
architecture as well as processing models of computational systems working
under control of resource managers;
- research on mechanisms discovering performance profiles of execution
of different classes of computational tasks on specific computer systems;
- preparing methods of computational and network resources state prediction.
|
