[Next]
[Up]
[Previous]
Next: Global Object Pointers
Up: A Tutorial Introduction to
Previous: Introduction
TACO implements a global memory for
Single Program Multiple Data (SPMD) programs
that consists of all combined local address spaces of the
participating processes.
TACO's objects are instances of arbitrary C++ classes and
the basic object model is therefore very lightweight and resource efficient.
Each process i can create and access all objects within
its own local space using standard C++ access methods.
However, creation of remote objects and accesses to
objects residing in other spaces can only be performed
by means of various synchronous and asynchronous remote
method invocation (RMI) techniques.
TACO is based on a notion of global object pointers
that are capable of addressing all objects uniquely in the global space.
Such global pointers are conceptually tuples consisting of
a reference to the local address space where
the object resides (usually a node or process number)
as well as a local pointer that points to a memory location
within that space. Consequently we can build dynamic
distributed data-structures with global pointers in
the same way as we can build dynamic data structures locally.
Such object groups can be addressed collectively
using collective methods.
Thus the basic object invocation paradigm is appropriately
extended to entire distributed object groups.
[Next]
[Up]
[Previous]
Next: Global Object Pointers
Up: A Tutorial Introduction to
Previous: Introduction