MultiCluster Project
Support Tools for Parallel Programming on Multiple Clusters
The MultiCluster project aims to define a model to efficiently
integrate different cluster-based architectures, such as Myrinet,
SCI and Fast Ethernet.
The main goal of the project is to identify
hardware and software aspects with the objective of develop a complete
programming environment to allow the user to configure such an integrated
architecture according to his application needs, in order to achieve high performance.
To accomplish such objectives, the MultiCluster project is structured
as follows:
a conceptual model regarding hardware and
software aspects with the insight of establish what kind of topology will be
considered as a multicluster, what kind of functionalities are expected from
the programming environment, how will be the application behaviour running in
this integrated platform, among others
development of DECK
(Distributed Execution and Communication Kernel), a runtime
kernel which is responsible to provide communication and
multiprogramming, as well as a set of specialized services (including
inter-cluster communication) to implement the concepts defined by the model
on top of Fast Ethernet, Myrinet and SCI-based clusters. The goal of DECK
is offer support to the development of parallel applications that will run
within a single cluster or over the multicluster architecture.
Besides these elements, it is possible to use DPC++
(Distributed Processing in C++), which is a distributed object-oriented programming
language running on top of DECK and furnishing intra-object concurrency,
load balancing and fault tolerance
facilities.