Scyld ClusterWare HPC

Administrator's Guide

Table of Contents

Preface

Feedback

Scyld ClusterWare Design Description

System Architecture

System Hardware Context
Network Topologies
System Data Flow
System Software Context
System Level Files

Technical Description

Compute Node Boot Procedure
BProc Distributed Process Space
Compute Node Categories
Compute Node States
Miscellaneous Components

Software Components

BeoBoot Tools
BProc Daemons
BProc Clients
ClusterWare Utilities

Configuring the Cluster with BeoSetup

BeoSetup Features

Starting BeoSetup

Full vs. Limited Privileges
Limiting Full Privileges

The BeoSetup Main Window

The File Menu

Configuration File...
Start Cluster
Service Reconfigure
Shutdown Cluster
Exit

The Settings Menu

Configuration
Preferences
Reread Cfg Files

The Help Menu

About...

The Toolbar

Apply
Revert
Node Floppy
Node CD
Config Boot
Configuration
Preferences

The Node List Boxes

The Configured Nodes List
The Unknown Addresses List
The Ignored Addresses List

Configuring the Cluster Manually

Configuration Files

/etc/beowulf/config
/etc/beowulf/fdisk
/etc/beowulf/fstab

Command Line Tools

bpstat
bpctl
node_down

The Kernel Command Line

Useful Command Line Options
Adding New Kernel Modules
Accessing External License Servers

Configuring RSH for Remote Job Execution

RSH from Master to Compute Node
RSH from Compute Node to Master

Configuring SSH for Remote Job Execution

Interconnects

Ethernet
Other Interconnects

Monitoring the Status of the Cluster

Monitoring Utilities

Cluster Monitoring Interfaces
Monitoring Daemons

Using the Data

BeoStatus

BeoStatus File Menu
BeoStatus Modes

beostat and libbeostat

bpstat

Managing Users on the Cluster

Managing User Accounts

Adding New Users
Removing Users

Managing User Groups

Creating a Group
Adding a User to a Group
Removing a Group

Controlling Access to Cluster Resources

What Node Ownership Means
Checking Node Ownership
Setting Node Ownership

Job Batching

Remote Administration and Monitoring

Command Line Tools
X Forwarding
Web-based Administration and Monitoring

Scyld Integrated Management Framework (IMF)

Working with the Web interface

Scyld IMF Components

Nodes
Torque Job and Queue Status
Ganglia
TaskMaster Portal
Settings
Help and Documentation

Advanced Customization for Developers

Architectural overview and data flow
Integrating existing Web interfaces
Adding diagnostic commands
Additional help

Compute Node Boot Options

Compute Node Boot Media

Floppy Disk
CD-ROM
PXE
Linux BIOS
Flash Disk

Changing Boot Settings

Adding Steps to the node_up Script
Per-Node Parameters
Other Per-Node Config Options

Error Logs

Disk Partitioning

Disk Partitioning Concepts

Disk Partitioning with ClusterWare

Master Node
Compute Nodes

Default Partitioning

Master Node
Compute Nodes

Partitioning Scenarios

Applying the Default Partitioning
Specifying Manual Partitioning

File Systems

File Systems on a Cluster

Local File Systems
Network/Cluster File Systems

NFS

NFS on Clusters
Configuration of NFS

ROMIO

Reasons to Use ROMIO
Installation and Configuration of ROMIO

Other Cluster File Systems

Supporting Multiple Master Nodes

Static Partitioning of Compute Nodes Among Multiple Masters
Cold Re-parenting of Compute Nodes
Node Migration Commands

Failover

When Compute Nodes Fail

Compute Node Data

When Master Nodes Fail

Protecting an Application from Node Failure

Preventing Node Failure

Load Balancing

Load Balancing in a Scyld Cluster

Mapping Policy
Queuing Policy

Implementing a Scheduling Policy

Extra Tools

TORQUE

Enabling TORQUE

IPMITool

Ganglia

Updating Software On Your Cluster

What Can't Be Updated

Special Directories, Configuration Files, and Scripts

What Resides on the Master Node

/etc/beowulf Directory
/usr/lib/beoboot Directory
/var/beowulf Directory
/var/log/beowulf Directory

What Gets Put on the Compute Nodes at Boot Time

Site-Local Startup Scripts

		Next >>
		Preface