Databases Reference
In-Depth Information
of handling different network conditions. Similarly, the process communica-
tion aspects of the distributed algorithms being proposed require significant
consideration.
A number of inter-process communication models have been developed in re-
cent years. These models include message-passing, shared memory, and mobile
agent. These models aid in the understanding of the communication proce-
dures occurring within a computational network. Message-passing is an inter-
process communication model developed as a guideline for process cooperation
between computing nodes in a parallel environment. This model is based on
a set of fundamental principles including:
1. Each process has its own local memory.
2. Processes communicate their data using a message exchange structure
(sending and receiving messages).
3. The transfer of data requires cooperative operations between each pro-
cess involved, i.e., each send operation must have a corresponding receive
operation.
The cooperative operations in the message-passing model address how com-
munication is being conducted between processing nodes within a network.
These operations form the components of a message-passing library, which
in return are used in the implementation of message-passing communication.
Examples of message-passing libraries include the Message Passing Library
(MPL) introduced for IBM SP2, the Parallel Virtual Machine (PVM), and
Message Passing Interface (MPI). The MPI library provides extensive porta-
bility and can be deployed on different types of platforms. The MPI specifica-
tion for message-passing has been established as a standard. An enhancement
of MPI standard, known as MPI-2, offers dynamic task control and a func-
tional parallel I/O capability [57]. Further discussions on this message-passing
model are included in Chapter 6.
