The DIVA library provides a transparent data management service for
computer systems in which several computers or processors are connected
by a relatively sparse network.
These networks are the typical interconnection in
large parallel and distributed systems such
as massively parallel processors (MPPs) and networks of workstations
(NOWs).
Each node of the network corresponds to a processor with its own memory
module of sufficiently large capacity and bandwidth such that the
communication links between
different processors are the major bottleneck of the system.
The DIVA library includes C-functions for creating and accessing global
variables, i.e., shared data objects.
The nodes can issue read and write requests that should be served using a
distributed data management strategy.
The most challenging algorithmic task is to use locality in order
to provide efficient access to the global variables.
The DIVA library implements a new generation of data management algorithms
that provably exploit the locality in the access pattern of a given
application as well as the locality in the topology of the underlying
network.
