Cluster Computing

Javier Delgado
Grid-Enabledment of Scientific Applications
Professor S. Masoud Sadjadi

Cluster Computing - GCB 1

Cluster Computing - GCB 2
Cluster Computing - GCB 3
 Essence of a Beowulf
 Hardware
 One head/master node
 (Several) compute nodes
 Interconnection modality (e.g. ethernet)
 Software
 Parallel Programming Infrastructure
 Scheduler (optional)
 Monitoring application (optional)

Cluster Computing - GCB 4

 Scheduling
 Multiple users fighting for resources = bad
 Don't allow them to do so directly
 Computer users are greedy
 Let the system allocate resources
 Users like to know job status without having to
keep an open session

Cluster Computing - GCB 5

 Cluster Solutions
 Do-it-yourself (DIY)
 OSCAR
 Rocks
 Pelican HPC (formerly Parallel Knoppix)
 Microsoft Windows CCE
 OpenMosix (closed March 2008)
 Clustermatic (no activity since 2005)

Cluster Computing - GCB 6

 DIY Cluster
 Advantages
 Control
 Learning Experience
 Disadvantages
 Control
 Administration

Cluster Computing - GCB 7

DIY-Cluster How-To Outline
 Hardware Requirements
 Head Node Deployment
 Core Software Requirements
 Cluster-specific Software
 Configuration
 Adding compute nodes

Cluster Computing - GCB 8

 Hardware Requirements
 Several commodity computers:
 cpu/motherboard
 memory
 ethernet card
 hard drive (recommended, in most cases)
 Network switch
 Cables, etc.

Cluster Computing - GCB 9

 Software Requirements –
Head node
 Core system
 system logger, core utilities, mail, etc.
 Linux Kernel
 Network Filesystem (NFS) server support
 Additional Packages
 Secure Shell (SSH) server
 iptables (firewall)
 nfs-utils
 portmap
 Network Time Protocol (NTP)
Cluster Computing - GCB 10
 Software Requirements –
Head node
 Additional Packages (cont.)
 inetd/xinetd – For FTP, globus, etc.
 Message Passing Interface (MPI) package
 Scheduler – PBS, SGE, Condor, etc.
 Ganglia – Simplified Cluster “Health” Logging
 dependency: Apache Web Server

Cluster Computing - GCB 11

 Initial Configuration
 Share /home directory
 Configure firewall rules
 Configure networking
 Configure SSH
 Create compute node image

Cluster Computing - GCB 12

 Building the Cluster
 Install compute node image on the compute
node
 Manually
 PXE Boot (pxelinux, etherboot, etc.)
 RedHat Kickstart
 etc.
 Configure host name, NFS, etc.
 ... for each node!

Cluster Computing - GCB 13

 Maintainance
 Software updates in head node require update
in compute node
 Failed nodes must be temporarily removed from
head node configuration files

Cluster Computing - GCB 14

 Building the Cluster
 But what if my boss wants a 200-node cluster?
 Monster.com
 OR come up with your own automation scheme
 OR Use OSCAR or Rocks

Cluster Computing - GCB 15

 Cluster Solutions
 Do-it-yourself (DIY)
 OSCAR
 Rocks
 Pelican HPC (formerly Parallel Knoppix)
 Microsoft Windows CCE
 OpenMosix (closed March 2008)
 Clustermatic (no activity since 2005)

Cluster Computing - GCB 16

 OSCAR
 Open Source Cluster Application Resources
 Fully-integrated software bundle to ease
deployment and management of a cluster
 Provides
 Management Wizard
 Command-line tools
 System Installation Suite

Cluster Computing - GCB 17

 Overview of Process
 Install OSCAR-approved Linux distribution
 Install OSCAR distribution
 Create node image(s)
 Add nodes
 Start computing

Cluster Computing - GCB 18

 OSCAR Management Wizard
 Download/install/remove OSCAR packages
 Build a cluster image
 Add/remove cluster nodes
 Configure networking
 Reimage or test a node with the Network Boot
Manager

Cluster Computing - GCB 19

 OSCAR Command Line tools
 Everything the Wizard offers
 yume
 Update node packages
 C3 - The Cluster Command Control Tools
 provide cluster-wide versions of common
commands
 Concurrent execution
 example 1: copy a file from the head node to all
visualization nodes
 example 2: execute a script on all compute nodes
Cluster Computing - GCB 20
 C3 List of Commands
 cexec: execution of any standard command on
all cluster nodes
 ckill: terminates a user specified process
 cget: retrieves files or directories from all cluster
nodes
 cpush: distribute files or directories to all cluster
nodes
 cpushimage: update the system image on all
cluster nodes using an image captured by the
SystemImager tool Cluster Computing - GCB 21
 List of Commands (cont.)
 crm: remove files or directories
 cshutdown: shutdown or restart all cluster
nodes
 cnum: returns a node range number based on
node name
 cname: returns node names based on node
ranges
 clist: returns all clusters and their type in a
configuration file
Cluster Computing - GCB 22
Example c3 configuration

# /etc/c3.conf
##
# describes cluster configuration
##
cluster gcb {
gcb.fiu.edu #head node
dead placeholder #change command line to 1 indexing
compute-0-[0-8] #first set of nodes
exclude 5 #offline node in the range (killed by J. Figueroa)
}
-------

Cluster Computing - GCB 23

 OPIUM
The OSCAR Password Installer and User


Management
 Synchronize user accounts
 Set up passwordless SSH

Periodically check for changes in passwords

Cluster Computing - GCB 24

 SIS
 System Installation Suite
 Installs Linux systems over a network
 Image-based
 Allows different images for different nodes
 Nodes can be booted from network, floppy, or
CD.

Cluster Computing - GCB 25

 Cluster Solutions
 Do-it-yourself (DIY)
 OSCAR
 Rocks
 Pelican HPC (formerly Parallel Knoppix)
 Microsoft Windows CCE
 OpenMosix (closed March 2008)
 Clustermatic (no activity since 2005)

Cluster Computing - GCB 26

 Rocks
 Disadvantages
 Tight-coupling of software
 Highly-automated
 Advantages
 Highly-automated...
 But also flexible

Cluster Computing - GCB 27

 Rocks
 The following 25 slides are property of UC
Regants

Cluster Computing - GCB 28

Cluster Computing - GCB 29
Cluster Computing - GCB 30
Cluster Computing - GCB 31
Cluster Computing - GCB 32
Cluster Computing - GCB 33
Cluster Computing - GCB 34
Cluster Computing - GCB 35
Cluster Computing - GCB 36
Cluster Computing - GCB 37
Cluster Computing - GCB 38
Cluster Computing - GCB 39
Cluster Computing - GCB 40
Cluster Computing - GCB 41
Cluster Computing - GCB 42
Cluster Computing - GCB 43
Cluster Computing - GCB 44
Determine number of nodes

Cluster Computing - GCB 45

Cluster Computing - GCB 46
Cluster Computing - GCB 47
Cluster Computing - GCB 48
Cluster Computing - GCB 49
Cluster Computing - GCB 50
Cluster Computing - GCB 51
 Rocks Installation
Simulation

Slides courtesy of David Villegas and Dany

Guevara

Cluster Computing - GCB 52

Cluster Computing - GCB 53
Cluster Computing - GCB 54
Cluster Computing - GCB 55
Cluster Computing - GCB 56
Cluster Computing - GCB 57
Cluster Computing - GCB 58
Cluster Computing - GCB 59
Cluster Computing - GCB 60
Cluster Computing - GCB 61
Cluster Computing - GCB 62
Cluster Computing - GCB 63
Cluster Computing - GCB 64
Cluster Computing - GCB 65
Cluster Computing - GCB 66
Cluster Computing - GCB 67
Installation of Compute Nodes


Log into Frontend node as root

At the command line run:
> insert-ethers

Cluster Computing - GCB 68

Cluster Computing - GCB 69
Cluster Computing - GCB 70
 Installation of Compute
Nodes
 Turn on the compute node
 Select to PXE boot or insert Rocks
CD and boot off of it

Cluster Computing - GCB 71

Cluster Computing - GCB 72
Cluster Computing - GCB 73
Cluster Computing - GCB 74
 Cluster Administration
 Command-line tools
 Image generation
 Cluster Troubleshooting
 User Management

Cluster Computing - GCB 75

 Command Line Tools
 Cluster-fork – execute command on nodes
(serially)
 Cluster-kill – kill a process on all nodes
 Cluster-probe – get information about cluster
status
 Cluster-ps – query nodes for a running process
by name

Cluster Computing - GCB 76

 Image Generation
 Basis: Redhat Kickstart file
 plus XML flexibility
 and dynamic stuff (i.e. support for “macros”)
 Image Location: /export/home/install
 Customization: rolls and extend-compute.xml
 Command: rocks-dist

Cluster Computing - GCB 77

 Image Generation

Cluster Computing - GCB 78

Source: http://www.rocksclusters.org/rocksapalooza/2007/dev-session1.pdf
 Example
 Goal: Make a regular node a visualization node
 Procedure
 Figure out what packages to install
 Determine what configuration files to modify
 Modify extend-compute.xml accordingly
 (Re-)deploy nodes

Cluster Computing - GCB 79

 Figure out Packages
 X-Windows Related
 X, fonts, display manager
 Display wall
 XDMX, Chromium, SAGE

Cluster Computing - GCB 80

 Modify Config Files
 X configuration
 xorg.conf
 Xinitrc
 Display Manager Configuration

Cluster Computing - GCB 81

 User Management
 Rocks Directory: /var/411
 Common configuration files:

Autofs-related
 /etc/group, /etc/passwd, /etc/shadow
 /etc/services, /etc/rpc
 All encrypted
 Helper Command
 rocks-user-sync

Cluster Computing - GCB 82

 Start Computing
 Rocks is now installed
 Choose an MPI runtime
 MPICH
 OpenMPI
 LAM-MPI
 Start compiling and executing

Cluster Computing - GCB 83

 Pelican HPC
 LiveCD for instant cluster creation
 Advantages
 Easy to use
 A lot of built-in software
 Disadvantages
 Not persistent
 Difficult to add software

Cluster Computing - GCB 84

 Microsoft Solutions
 Windows Server 2003 Compute Cluster Edition
(CCE)
 Microsoft Compute Cluster pack (CCP)
 Microsoft MPI (based on MPICH2)
 Microsoft Scheduler

Cluster Computing - GCB 85

 Microsoft CCE
 Advantages
 Using Remote Installation Services (RIS), compute
nodes can be added by simply turning it on
 May be better for those familiar with Microsoft
Environment
 Disadvantages
 Expensive
 Only for 64-bit architectures
 Proprietary
 Limited Application base
Cluster Computing - GCB 86
 References
 http://pareto.uab.es/mcreel/PelicanHPC/
 http://pareto.uab.es/mcreel/ParallelKnoppix/
 http://www.gentoo.org/doc/en/hpc-howto.xml
 http://www.clustermatic.org

http://www.microsoft.com/windowsserver2003/c
cs/default.aspx
 http://www.redhat.com/docs/manuals/linux/RHL
-9-Manual/ref-guide/ch-nfs.html
 portmap man page
Cluster Computing - GCB 87
 http://www.rocksclusters.org/rocksapalooza