Home > Articles > Operating Systems, Server > Solaris

  • Print
  • + Share This
Like this article? We recommend

Beowulf Solution

Beowulf is one way to build a compute cluster. A Beowulf cluster is configured with PC hardware running the Linux operating system and other open source software. Often, standard Fast Ethernet is used for the interconnect, but sometimes Gigabit Ethernet or Myrinet are used. The file sharing is NFS and the compilers are GNU. If parallel jobs are to be processed, the open source MPICH MPI software implementation is used. For batch queuing, PBS is the most commonly used software. Because the clusters use inexpensive PC hardware and free open source software, the initial cost for building a Beowulf cluster is very low. Note, however, that Solaris OE and SPARC_ product combinations that cost less than $1,000 are available, too. So the perception that only a cluster based on PC hardware is inexpensive is incorrect. The drawbacks of using the PC solution are that you must support the solution yourself and most PC hardware is still 32 bit, so you cannot run any applications that require much memory. The current PC hardware and the Linux operating system cannot handle many CPUs efficiently in an SMP environment. This limitation means that the traditional Beowulf implementation of a compute cluster only fits the low end of the computing spectrum, plus some parallel applications that also scale well over a slow interconnect. TABLE 2 summarizes the typical PC-based Beowulf components.

TABLE 2 PC-Based Beowulf Cluster Components

Service

Component

Architecture

X86

Operating System

Linux

Network

Ethernet or Myricom Myrinet

File sharing

NFS

Queuing system

PBS or Sun Grid Engine software

Message passing interface

MPICH

Compiler

GNU

Management tools—job management

PBS or Sun Grid Engine software


  • + Share This
  • 🔖 Save To Your Account