Scalable Remote Sensing Applications Development on Beowulf

The latest system software works to implement a Beowulf aware environment that presents a single system image to the user. No longer simply a collection of computers, a Beowulf is a computing system with a collection of processing resources.

• bproc - process management; Processes can be started on any node and managed from a central location

• bnm/bMPI - message passing; Beowulf low-level message passing layers and standard user-level interface

• balloc - processor allocation; Manages allocation of processors to jobs

• PVFS - parallel file system; Provides a global file space and parallel file I/O for high performance

• bmon - monitoring; A unified system for gathering data on events and performance at system and user job levels

• bdb - debugging Symbolic debugging aware of bproc, bMPI, PVFS; and utilizing bmon

Beowulf system software consists of interoperable and replaceable modules that encourage reuse, innovation, and growth.

Additional Information:

Parallel Architecture Research Laboratory (PARL)

The Parallel Virtual File System provides high performance parallel file access for Linux clusters. It also serves as a research platform for further study into parallel I/O.

Features of PVFS:

• Provides high bandwidth for concurrent read and write operations from multiple processes to a common file

• Supports multiple APIs: a native PVFS API, the POSIX I/O API, and MPI-IO support through ROMIO

• Compatibility with common existing Unix tools without recompilation

• Scalability and robustness

• Easy to install and use

Current Research Topics in PVFS:

• Reactive scheduling: adapting I/O algorithms dynamically to reflect workload and system characteristics

• Alternative network communications: ability to support data transfers using methods other than TCP/IP

• Alternative disk access: ability to access local I/O node data through various mechanisms such as raw I/O or asynchronous I/O

• Advanced API: more efficient expression of parallel file access requests and higher performance UNIX compatibility

The Parallel Virtual File System

The Problem Solving Environment (PSE) effort seeks to improve the accessibility of Beowulf clusters and other forms of high performance computing devices. CECAAD is an infrastructure and toolkit for the creation of PSEs.

• ADF - a dynamic open format for representing shared designs

• An Agent Model for creating a set of cooperating tools with appropriate abstractions

• A Manager and Library for managing concurrent access to designs

• A layered abstraction model

• A set of common agents and agent components

RCADE is a CECAAD-based environment targeted at creating remote sensing applications for FPGA-based reconfigurable computing devices.

Graphical design entry
A library of optimized preplaced components
VHDL or JHDL code generation
Precision and throughput analysis
Automatic pipeline balancing and buffer insertion

Additional Information:

Problem Solving Environments

CERSe provides an environment for design, testing, and production execution of remote-sensing algorithms. With CERSe, programmers reuse existing remote-sensing code to build working parallel applications.

• Modules - plug-compatible components;
  Complex algorithms are constructed from building blocks of smaller algorithms

• Automatic Parallelism - high performance
  Modules are allocated pieces of a dataset to process and can choose to ignore global issues

• Dataflow Manager - localized management
  Handles data partitioning and distribution as well as routing and coordination of processing agents

• TPH - Tagged Partition Handle
  Content-addressable data structure for passing data between modules

• Interface Description Language - easy module design
  Modules are written using a CORBA-like interface and seamlessly incorporated into a CERSe program

• GUI - development and simulation environment
  An intuitive GUI for development and execution of algorithms with tools for visualization and performance analysis

High performance, usability, and extensibility are key features provided by CERSe. This environment takes advantage of high performance computers yet allows easy development so that users can focus on their domain science instead of programming.

Additional Information:

Parallel Architecture Research Laboratory

Parallel Virtual File System Home Page

Problem Solving Environments Home Page

CERSe Home Page

Point of Contact:
Walt Ligon
Clemson University
864-656-1224
walt@clemson.edu